feat(opencode): complete OpenCode agent setup - add 11 missing agent prompts#726
Open
nayanjaiswal1 wants to merge 2 commits intoaffaan-m:mainfrom
Open
feat(opencode): complete OpenCode agent setup - add 11 missing agent prompts#726nayanjaiswal1 wants to merge 2 commits intoaffaan-m:mainfrom
nayanjaiswal1 wants to merge 2 commits intoaffaan-m:mainfrom
Conversation
Contributor
|
Contributor
Analysis Failed
Troubleshooting
Retry: |
Contributor
📝 WalkthroughWalkthroughAdds 13 new OpenCode agent registrations and matching prompt/config files, covering language-specific reviewers and build resolvers, a docs-lookup tool, and workflow/orchestration agents (chief-of-staff, loop-operator, harness-optimizer). Changes
Estimated code review effort🎯 3 (Moderate) | ⏱️ ~20 minutes Possibly related PRs
Suggested reviewers
Poem
🚥 Pre-merge checks | ✅ 3✅ Passed checks (3 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
Contributor
Greptile SummaryThis PR completes the OpenCode agent setup by adding 11 new prompt files (chief-of-staff, cpp-reviewer, cpp-build-resolver, docs-lookup, harness-optimizer, java-reviewer, java-build-resolver, kotlin-reviewer, kotlin-build-resolver, loop-operator, python-reviewer), the companion
Confidence Score: 3/5
Important Files Changed
Flowchart%%{init: {'theme': 'neutral'}}%%
flowchart TD
User([User Request]) --> OC[OpenCode Runtime]
OC -->|loads| AGENTS[AGENTS.md instruction]
OC -->|dispatches| AgentRouter{Agent Router}
AgentRouter --> ReviewAgents[Reviewer Agents]
AgentRouter --> BuildAgents[Build Resolver Agents]
AgentRouter --> UtilityAgents[Utility Agents]
ReviewAgents --> CPP[cpp-reviewer\nread+bash only]
ReviewAgents --> Java[java-reviewer\nread+bash only]
ReviewAgents --> Kotlin[kotlin-reviewer\nread+bash only]
ReviewAgents --> Python[python-reviewer\nread+bash only]
BuildAgents --> CPPB[cpp-build-resolver\nread+write+edit+bash]
BuildAgents --> JavaB[java-build-resolver\nread+write+edit+bash]
BuildAgents --> KotlinB[kotlin-build-resolver\nread+write+edit+bash]
UtilityAgents --> CoS[chief-of-staff\nread+write+edit+bash]
UtilityAgents --> Docs[docs-lookup\nread+bash only]
UtilityAgents --> HO[harness-optimizer\nread+bash+edit\n⚠️ write missing]
UtilityAgents --> LO[loop-operator\nread+bash+edit\n⚠️ write missing]
CoS -->|reads| SOUL[SOUL.md\ntone context]
AGENTS -->|advertises ⚠️| Ghost1[pytorch-build-resolver\nno prompt / not in JSON]
AGENTS -->|advertises ⚠️| Ghost2[typescript-reviewer\nno prompt / not in JSON]
|
![greptile-apps[bot]](https://avatars.githubusercontent.com/in/867647?s=60&v=4){kind=small}
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
Contributor
There was a problem hiding this comment.
Actionable comments posted: 8
🧹 Nitpick comments (5)
.opencode/prompts/agents/harness-optimizer.txt (1)
24-27: Align output fields with/harness-auditcontract
baseline scorecardis ambiguous against the command’s deterministic output (overall_score, category scores,top_actions). Please mirror those keys to keep downstream parsing stable.Suggested output wording update
- - baseline scorecard + - baseline: overall_score/max_score + category scores + top_actions - applied changes - measured improvements - remaining risks🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In @.opencode/prompts/agents/harness-optimizer.txt around lines 24 - 27, Replace the ambiguous output fields in .opencode/prompts/agents/harness-optimizer.txt so they mirror the /harness-audit contract: change the string "baseline scorecard" to use the deterministic keys like "overall_score" (and include category-specific score keys e.g., "security_score", "cost_score" as applicable), map "applied changes" to "top_actions" (or an array of action objects), ensure "measured improvements" emit explicit category score deltas using the same category keys, and keep "remaining risks" as a clear "remaining_risks" field; update any surrounding wording to emit those exact keys so downstream parsing matches the /harness-audit schema..opencode/prompts/agents/python-reviewer.txt (1)
61-61: Avoid repo-specific coverage target in default diagnostics.
--cov=appassumes a package name that may not exist. Prefer a generic command or explicit placeholder.Suggested prompt adjustment
-pytest --cov=app --cov-report=term-missing # Test coverage +pytest --cov --cov-report=term-missing # Test coverage (or replace with project package)🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In @.opencode/prompts/agents/python-reviewer.txt at line 61, Replace the repo-specific pytest invocation that contains "--cov=app" (the line "pytest --cov=app --cov-report=term-missing") with a generic or placeholder coverage flag; update the prompt to use either no --cov option or a generic token like "--cov=<package>" or "pytest --cov=<PACKAGE> --cov-report=term-missing" so the default diagnostics do not assume a package named "app"..opencode/prompts/agents/java-reviewer.txt (1)
64-70: Align diagnostic commands with detected build tool.The prompt says to determine build tool first, but the command list still mixes Maven/Gradle goals unconditionally. Split the diagnostics into “Maven-only” and “Gradle-only” sections to reduce false failures.
Also applies to: 74-74
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In @.opencode/prompts/agents/java-reviewer.txt around lines 64 - 70, The command list currently runs both Maven and Gradle goals unconditionally; update the diagnostics in the prompt to split commands into Maven-only and Gradle-only groups and invoke only the appropriate group based on the earlier detected build tool (e.g., using the existing detection result for pom.xml vs build.gradle). Specifically, replace the mixed sequence (mvn verify, ./gradlew check, ./mvnw checkstyle:check, ./mvnw spotbugs:check, ./mvnw test, ./mvnw dependency-check:check, grep ...) with two clearly labeled lists and ensure the agent executes only the matching list for Maven or Gradle, and apply the same change to the similar block referenced at lines 74-74..opencode/prompts/agents/kotlin-build-resolver.txt (2)
57-57: Avoid destructive cache deletion in baseline troubleshooting flow.
rm -rf .gradle/build-cache/is heavy for a default diagnostic step and can slow iteration significantly. Keep this as a last-resort note, not a standard command.Based on learnings: Use build-error-resolver agent to analyze errors, fix incrementally, and verify after each fix.Suggested prompt adjustment
-# Clear project-local Gradle build cache -./gradlew clean && rm -rf .gradle/build-cache/ +# Clean build outputs (use cache deletion only as last resort) +./gradlew clean🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In @.opencode/prompts/agents/kotlin-build-resolver.txt at line 57, The diagnostic step currently runs "./gradlew clean && rm -rf .gradle/build-cache/" which dangerously deletes the Gradle build cache by default; change the prompt in kotlin-build-resolver.txt so the default command is just "./gradlew clean" (remove "&& rm -rf .gradle/build-cache/"), and instead add a short follow-up note that "rm -rf .gradle/build-cache/" is a last-resort action to use only after incremental fixes fail; also reference using the build-error-resolver agent to analyze errors, apply fixes incrementally, and verify after each change (mention build-error-resolver and the cleaned "./gradlew clean" command to locate where to update).
73-77: Include Kotlin/Gradle version compatibility notes forcompilerOptionssyntax.The
compilerOptionsblock in the example requires Kotlin Gradle Plugin 1.8.0 or later. Add a note indicating the version requirement and fallback syntax for older projects:Suggested addition after the code block
Note: compilerOptions syntax requires KGP 1.8.0+. For older versions (KGP < 1.8.0), use: tasks.withType(org.jetbrains.kotlin.gradle.tasks.KotlinCompile::class.java).configureEach { kotlinOptions { jvmTarget = "17" freeCompilerArgs += listOf("-Xjsr305=strict") allWarningsAsErrors = true } }This helps the agent handle builds with varying Kotlin versions gracefully.
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In @.opencode/prompts/agents/kotlin-build-resolver.txt around lines 73 - 77, Add a compatibility note after the kotlin { compilerOptions { ... } } block that states the compilerOptions syntax requires Kotlin Gradle Plugin (KGP) 1.8.0 or newer and provide a fallback approach for older KGP versions (KGP < 1.8.0) using tasks.withType(org.jetbrains.kotlin.gradle.tasks.KotlinCompile::class.java).configureEach to set kotlinOptions (including jvmTarget, freeCompilerArgs and allWarningsAsErrors) so agents working with older projects can apply the equivalent settings.
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Inline comments:
In @.opencode/opencode.json:
- Around line 217-225: The docs-lookup agent entry currently grants only "read"
and "bash" tools but needs explicit Context7 MCP tool permissions; update the
"docs-lookup" object (the agent named "docs-lookup") to either add a
"permission" block granting "context7_*" tool access (e.g.,
"permission":{"tool":{"context7_*":"allow"}}) or add legacy explicit tool flags
for the required Context7 actions (e.g., include "context7_resolve-library-id":
true and "context7_query-docs": true under "tools") so the agent can call
resolve-library-id and query-docs without runtime permission prompts.
In @.opencode/prompts/agents/chief-of-staff.txt:
- Around line 69-77: The step "Git commit & push — Version-control all knowledge
file changes" in the post-send checklist introduces automatic external side
effects; change it to a safe, non-automatic workflow by replacing that line with
a queued/manual flow (e.g., "Git: Stage changes and queue for review; require
explicit approval to commit & push" or "Create local commit only; do NOT push
without approval"), and add a note that pushing requires explicit user
confirmation or an opt-in flag; update any related consumer text that assumes
auto-push behavior to reflect the new manual/queued commit flow.
In @.opencode/prompts/agents/cpp-build-resolver.txt:
- Around line 13-18: The current command sequence runs "cmake --build build"
before generating build files and may fail on clean environments; reorder the
steps so you run "cmake -B build -S ." (configure/generate) before "cmake
--build build" and keep the diagnostics lines (the tail/head wrappers and
"clang-tidy src/*.cpp -- -std=c++17") so the build-resolver agent config runs
configure first, then build, then static analysis to allow incremental
troubleshooting with the build-error-resolver agent.
In @.opencode/prompts/agents/java-build-resolver.txt:
- Line 44: The table row currently uses the British spelling "Initialise
variable before use" while nearby text uses US spelling ("Uninitialized");
update that cell to the US variant by changing "Initialise variable before use"
to "Initialize variable before use" so the wording is consistent across the
prompt.
- Around line 15-25: The prompt currently runs both Maven and Gradle commands
unconditionally; change it to detect the repo's build system first (look for
mvnw/mvn or pom.xml to pick Maven, gradlew/gradle or build.gradle(.kts) to pick
Gradle) and only execute the appropriate command set; for Maven-run the compile,
test, dependency:tree (and checkstyle/spotbugs fallbacks) lines currently shown
for mvnw/mvn, and for Gradle-run the build and runtimeClasspath dependency query
lines for gradlew/gradle, ensuring the script uses the detected wrapper (mvnw vs
mvn, gradlew vs gradle) and skips the other tool entirely to avoid noise.
In @.opencode/prompts/agents/loop-operator.txt:
- Around line 15-20: The "Required Checks" list names mandatory safety
conditions (quality gates, eval baseline, rollback path, branch/worktree
isolation) but the prompt never enforces them; update the loop operator prompt
to add an explicit pre-execution validation step that verifies each check and
aborts the loop if any are missing or failing (e.g., "Before starting the loop,
confirm quality gates pass, an eval baseline exists, a rollback path is
available, and branch/worktree isolation is configured; if any check fails, stop
and report which check failed"). Reference the "Required Checks" section and
ensure the prompt language requires the agent to actively validate and report
these four items before proceeding.
In @.opencode/prompts/agents/python-reviewer.txt:
- Around line 14-15: The checklist's Path Traversal item is insufficient:
replace the current "normpath + '..' filtering" guidance with a requirement to
canonicalize user-supplied paths (e.g., using realpath/resolve) and validate
them against a base-directory allowlist via a secure commonpath/startsWith
check, explicitly reject absolute paths or paths that escape the base via
symlinks, and add a note to test with symlink/absolute-path bypass cases; update
the prompt text that mentions normpath and `..` filtering to reference
canonicalization and base-directory allowlisting as mandatory controls.
---
Nitpick comments:
In @.opencode/prompts/agents/harness-optimizer.txt:
- Around line 24-27: Replace the ambiguous output fields in
.opencode/prompts/agents/harness-optimizer.txt so they mirror the /harness-audit
contract: change the string "baseline scorecard" to use the deterministic keys
like "overall_score" (and include category-specific score keys e.g.,
"security_score", "cost_score" as applicable), map "applied changes" to
"top_actions" (or an array of action objects), ensure "measured improvements"
emit explicit category score deltas using the same category keys, and keep
"remaining risks" as a clear "remaining_risks" field; update any surrounding
wording to emit those exact keys so downstream parsing matches the
/harness-audit schema.
In @.opencode/prompts/agents/java-reviewer.txt:
- Around line 64-70: The command list currently runs both Maven and Gradle goals
unconditionally; update the diagnostics in the prompt to split commands into
Maven-only and Gradle-only groups and invoke only the appropriate group based on
the earlier detected build tool (e.g., using the existing detection result for
pom.xml vs build.gradle). Specifically, replace the mixed sequence (mvn verify,
./gradlew check, ./mvnw checkstyle:check, ./mvnw spotbugs:check, ./mvnw test,
./mvnw dependency-check:check, grep ...) with two clearly labeled lists and
ensure the agent executes only the matching list for Maven or Gradle, and apply
the same change to the similar block referenced at lines 74-74.
In @.opencode/prompts/agents/kotlin-build-resolver.txt:
- Line 57: The diagnostic step currently runs "./gradlew clean && rm -rf
.gradle/build-cache/" which dangerously deletes the Gradle build cache by
default; change the prompt in kotlin-build-resolver.txt so the default command
is just "./gradlew clean" (remove "&& rm -rf .gradle/build-cache/"), and instead
add a short follow-up note that "rm -rf .gradle/build-cache/" is a last-resort
action to use only after incremental fixes fail; also reference using the
build-error-resolver agent to analyze errors, apply fixes incrementally, and
verify after each change (mention build-error-resolver and the cleaned
"./gradlew clean" command to locate where to update).
- Around line 73-77: Add a compatibility note after the kotlin { compilerOptions
{ ... } } block that states the compilerOptions syntax requires Kotlin Gradle
Plugin (KGP) 1.8.0 or newer and provide a fallback approach for older KGP
versions (KGP < 1.8.0) using
tasks.withType(org.jetbrains.kotlin.gradle.tasks.KotlinCompile::class.java).configureEach
to set kotlinOptions (including jvmTarget, freeCompilerArgs and
allWarningsAsErrors) so agents working with older projects can apply the
equivalent settings.
In @.opencode/prompts/agents/python-reviewer.txt:
- Line 61: Replace the repo-specific pytest invocation that contains "--cov=app"
(the line "pytest --cov=app --cov-report=term-missing") with a generic or
placeholder coverage flag; update the prompt to use either no --cov option or a
generic token like "--cov=<package>" or "pytest --cov=<PACKAGE>
--cov-report=term-missing" so the default diagnostics do not assume a package
named "app".
ℹ️ Review info
⚙️ Run configuration
Configuration used: defaults
Review profile: CHILL
Plan: Pro
Run ID: 402c763a-81f1-448c-b468-dc1fb5c46348
📒 Files selected for processing (12)
.opencode/opencode.json.opencode/prompts/agents/chief-of-staff.txt.opencode/prompts/agents/cpp-build-resolver.txt.opencode/prompts/agents/cpp-reviewer.txt.opencode/prompts/agents/docs-lookup.txt.opencode/prompts/agents/harness-optimizer.txt.opencode/prompts/agents/java-build-resolver.txt.opencode/prompts/agents/java-reviewer.txt.opencode/prompts/agents/kotlin-build-resolver.txt.opencode/prompts/agents/kotlin-reviewer.txt.opencode/prompts/agents/loop-operator.txt.opencode/prompts/agents/python-reviewer.txt
Comment on lines
+217
to
+225
| "docs-lookup": { | ||
| "description": "Documentation specialist using Context7 MCP to fetch current library and API documentation with code examples.", | ||
| "mode": "subagent", | ||
| "model": "anthropic/claude-sonnet-4-5", | ||
| "prompt": "{file:prompts/agents/docs-lookup.txt}", | ||
| "tools": { | ||
| "read": true, | ||
| "bash": true | ||
| } |
Contributor
There was a problem hiding this comment.
🧩 Analysis chain
🏁 Script executed:
#!/bin/bash
# Verify current wiring and whether any agent explicitly grants MCP tools
rg -n '"docs-lookup"|mcp_\*|"tools"|context7' .opencode/opencode.json mcp-configs/mcp-servers.json
rg -n '"mcp"' .opencode/opencode.jsonRepository: affaan-m/everything-claude-code
Length of output: 1470
🌐 Web query:
In OpenCode agent configuration, what is the exact way to grant a subagent access to MCP tools (e.g., Context7 resolve-library-id/query-docs)? Is a dedicated mcptool flag required underagent..tools?
💡 Result:
To grant a subagent access to MCP tools like Context7's resolve-library-id or query-docs in OpenCode agent configuration, configure the subagent's tools or permissions under agent.<name>. Explicitly enable the relevant MCP-prefixed tools using glob patterns. MCP tools from a server named "context7" appear as "context7_" (e.g., "context7_resolve-library-id"). First, define the MCP server globally in the top-level "mcp" section of opencode.json: { "mcp": { "context7": { "type": "remote", "url": "https://mcp.context7.com/mcp" } } } Then, for the subagent, enable access in agent.<name>.tools (legacy but supported) or preferably agent.<name>.permission.tool: Using tools (legacy/deprecated): { "agent": { "my-subagent": { "mode": "subagent", "tools": { "context7_": true } } } } Using permissions (recommended, new configs): { "agent": { "my-subagent": { "mode": "subagent", "permission": { "tool": { "context7_": "allow" } } } } } A dedicated "mcp" tool flag is not required; MCP tools are treated as regular tools with servername_ naming and controlled via tools/permissions like built-ins. Globally disable if needed ("tools": {"context7_*": false}), then override per-agent. Note: There is a known bug (as of 2026-03-07) where task-spawned subagents may not inherit MCP permissions properly despite tools appearing available; explicitly configure agent permissions to mitigate.
Citations:
- 1: https://opencodeguide.com/en/docs/configure/agents/
- 2: https://opencode.ai/docs/agents/
- 3: https://opencode.ai/docs/tools/
- 4: https://dev.opencode.ai/docs/agents/
- 5: [BUG] Subagents spawned via Task tool cannot execute MCP tools despite appearing in tool registry anomalyco/opencode#16491
- 6: https://dev.opencode.ai/docs/mcp-servers/
- 7: https://opencode.ai/docs/mcp-servers/
🏁 Script executed:
# Read the docs-lookup prompt file to verify it requires Context7 MCP tools
find . -name "docs-lookup.txt" -o -name "prompts" -type d | head -5Repository: affaan-m/everything-claude-code
Length of output: 137
🏁 Script executed:
# Check if there are any agents that DO explicitly grant context7 tools
rg -A 5 '"tools"\s*:\s*\{' .opencode/opencode.json | grep -A 5 'context7'Repository: affaan-m/everything-claude-code
Length of output: 57
🏁 Script executed:
# Look at the full docs-lookup agent config and nearby agents for comparison
sed -n '217,230p' .opencode/opencode.jsonRepository: affaan-m/everything-claude-code
Length of output: 654
🏁 Script executed:
# Read the docs-lookup prompt to verify it requires Context7 MCP
cat ./.opencode/prompts/agents/docs-lookup.txtRepository: affaan-m/everything-claude-code
Length of output: 2883
🏁 Script executed:
# Search for any agents that DO grant context7 tools to see proper pattern
rg -B 2 'context7' .opencode/opencode.jsonRepository: affaan-m/everything-claude-code
Length of output: 57
Grant docs-lookup explicit access to Context7 MCP tools
The agent's prompt explicitly requires resolve-library-id and query-docs from Context7 MCP, but the config only grants read and bash. Without explicit tool grants, the agent will trigger permission requests at runtime instead of executing deterministically.
Add to the docs-lookup agent:
"permission": {
"tool": {
"context7_*": "allow"
}
}Or use the legacy syntax: "tools": {"context7_resolve-library-id": true, "context7_query-docs": true}
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In @.opencode/opencode.json around lines 217 - 225, The docs-lookup agent entry
currently grants only "read" and "bash" tools but needs explicit Context7 MCP
tool permissions; update the "docs-lookup" object (the agent named
"docs-lookup") to either add a "permission" block granting "context7_*" tool
access (e.g., "permission":{"tool":{"context7_*":"allow"}}) or add legacy
explicit tool flags for the required Context7 actions (e.g., include
"context7_resolve-library-id": true and "context7_query-docs": true under
"tools") so the agent can call resolve-library-id and query-docs without runtime
permission prompts.
Comment on lines
+14
to
+15
| - **Path Traversal**: user-controlled paths — validate with normpath, reject `..` | ||
| - **Eval/exec abuse**, **unsafe deserialization**, **hardcoded secrets** |
Contributor
There was a problem hiding this comment.
Path traversal guidance is incomplete for CRITICAL security checks.
normpath + .. filtering alone can still be bypassed (absolute paths/symlinks). This should require canonicalization and base-directory allowlisting in the checklist.
Suggested prompt adjustment
-- **Path Traversal**: user-controlled paths — validate with normpath, reject `..`
+- **Path Traversal**: user-controlled paths — resolve to canonical path and enforce it stays within an allowed base directory (do not rely only on `..` checks)🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In @.opencode/prompts/agents/python-reviewer.txt around lines 14 - 15, The
checklist's Path Traversal item is insufficient: replace the current "normpath +
'..' filtering" guidance with a requirement to canonicalize user-supplied paths
(e.g., using realpath/resolve) and validate them against a base-directory
allowlist via a secure commonpath/startsWith check, explicitly reject absolute
paths or paths that escape the base via symlinks, and add a note to test with
symlink/absolute-path bypass cases; update the prompt text that mentions
normpath and `..` filtering to reference canonicalization and base-directory
allowlisting as mandatory controls.
dragosnastasemihai-design
approved these changes
Mar 21, 2026
Contributor
|
Contributor
Analysis Failed
Troubleshooting
Retry: |
Author
|
Thank you for the detailed review! I've addressed all the issues in commit : Greptile Critical Issues Fixed:
CodeRabbit Nitpicks Fixed:
All fixes are pushed to . |
nayanjaiswal1
commented
Mar 21, 2026
Author
nayanjaiswal1
left a comment
nayanjaiswal1
left a comment
There was a problem hiding this comment.
Thanks @coderabbitai! All nitpick suggestions have been addressed in commit :
- : Output fields now match /harness-audit contract with proper score keys
- : Generic pytest coverage command
- : Split Maven/Gradle diagnostics
- : Removed aggressive cache deletion + added KGP version fallback
- : Fixed cmake command order
- : Fixed spelling + added build detection
- : Added pre-execution validation step
- : Manual git push flow
Author
|
Note: The Path Traversal suggestion for python-reviewer.txt (line 14-15 about canonicalization vs normpath) was reviewed but not addressed in this PR — the current normpath guidance remains. |
Contributor
There was a problem hiding this comment.
Actionable comments posted: 3
🧹 Nitpick comments (3)
.opencode/prompts/agents/SOUL.md (1)
45-47: Use consistent minute formatting for readability (30 minvs30min).Not blocking, but spacing units improves doc clarity and avoids grammar-lint noise.
Proposed fix
-- **Default meeting duration**: 30min / 45min / 60min (default: 30min) +- **Default meeting duration**: 30 min / 45 min / 60 min (default: 30 min) ... -- **Buffer between meetings**: 0 / 5 / 15 min (default: 5min) +- **Buffer between meetings**: 0 / 5 / 15 min (default: 5 min)🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In @.opencode/prompts/agents/SOUL.md around lines 45 - 47, Update the minute formatting in the listed options to include a space between the number and "min" for consistency: change "30min", "45min", "60min" to "30 min", "45 min", "60 min" in the "Default meeting duration" line, and change "5min" to "5 min" in the "Buffer between meetings" line; keep the rest of the wording identical..opencode/prompts/agents/java-reviewer.txt (1)
68-69: Avoid|| mvn ...fallback that can mask real Maven wrapper failures.At Line 68 and Line 69,
./mvnw ... || mvn ...will runmvnafter any wrapper command failure (including real compile/test failures), which can obscure diagnostics. Prefer explicit wrapper-exists branching.🔧 Proposed update
-./mvnw compile -q 2>&1 || mvn compile -q 2>&1 -./mvnw verify -q 2>&1 || mvn verify -q 2>&1 +if [ -x ./mvnw ]; then ./mvnw compile -q 2>&1; else mvn compile -q 2>&1; fi +if [ -x ./mvnw ]; then ./mvnw verify -q 2>&1; else mvn verify -q 2>&1; fi🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In @.opencode/prompts/agents/java-reviewer.txt around lines 68 - 69, The current fallback "./mvnw ... || mvn ..." masks genuine wrapper failures by always running mvn after any mvnw failure; replace the fallback with an explicit check for the wrapper’s presence/executability and run either the wrapper or the system maven accordingly for both the compile and verify steps (identify occurrences of "./mvnw compile -q" and "./mvnw verify -q" in the file), so only when the wrapper is truly absent/unusable do we run "mvn" and real wrapper errors aren’t hidden..opencode/prompts/agents/loop-operator.txt (1)
26-31: Consider removing redundant Required Checks section.Lines 26-31 duplicate the checks already listed in the Pre-Execution Validation section (lines 19-22). Since the validation section includes both the checks AND enforcement instructions, this section adds redundancy without additional value.
♻️ Proposed simplification
If any check fails, **STOP immediately** and report which check failed before proceeding. -## Required Checks - -- quality gates are active -- eval baseline exists -- rollback path exists -- branch/worktree isolation is configured - ## Escalation🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In @.opencode/prompts/agents/loop-operator.txt around lines 26 - 31, Remove the redundant "## Required Checks" block (the four bullet items: "quality gates are active", "eval baseline exists", "rollback path exists", "branch/worktree isolation is configured") since those checks are already covered under the "Pre-Execution Validation" section; locate the "Required Checks" heading and delete that heading plus its bullet list so the document only keeps the singular "Pre-Execution Validation" entry.
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Inline comments:
In @.opencode/prompts/agents/java-reviewer.txt:
- Line 4: Replace the context-sensitive git invocation "git diff -- '*.java'"
with a PR-aware diff that compares the feature branch against the target
baseline (for example origin/main...HEAD) and restricts output to Java files
(use the name-only output and the '*.java' pathspec); update the command in the
prompt line so the reviewer reliably sees all changed Java files across
clean/sandboxed checkouts.
In @.opencode/prompts/agents/SOUL.md:
- Line 12: Replace the malformed bold key `** formality_level**` with a
correctly formatted bold key by removing the leading space so it reads
`**formality_level**`; update the instance in SOUL.md (the `formality_level`
key) to match the other bolded keys' formatting.
- Around line 1-4: SOUL.md is missing the required YAML frontmatter which breaks
agent discovery; add a top-of-file YAML frontmatter block that declares the
agent's name, a short description, the tools array, and the model string (i.e.,
include keys: name, description, tools, model) so the agent parser can recognize
it; ensure the frontmatter is valid YAML (surrounded by ---) and that the tools
field lists any tool identifiers used by this agent and the model field
specifies the model to use, then save SOUL.md with the frontmatter followed by
the existing Markdown content.
---
Nitpick comments:
In @.opencode/prompts/agents/java-reviewer.txt:
- Around line 68-69: The current fallback "./mvnw ... || mvn ..." masks genuine
wrapper failures by always running mvn after any mvnw failure; replace the
fallback with an explicit check for the wrapper’s presence/executability and run
either the wrapper or the system maven accordingly for both the compile and
verify steps (identify occurrences of "./mvnw compile -q" and "./mvnw verify -q"
in the file), so only when the wrapper is truly absent/unusable do we run "mvn"
and real wrapper errors aren’t hidden.
In @.opencode/prompts/agents/loop-operator.txt:
- Around line 26-31: Remove the redundant "## Required Checks" block (the four
bullet items: "quality gates are active", "eval baseline exists", "rollback path
exists", "branch/worktree isolation is configured") since those checks are
already covered under the "Pre-Execution Validation" section; locate the
"Required Checks" heading and delete that heading plus its bullet list so the
document only keeps the singular "Pre-Execution Validation" entry.
In @.opencode/prompts/agents/SOUL.md:
- Around line 45-47: Update the minute formatting in the listed options to
include a space between the number and "min" for consistency: change "30min",
"45min", "60min" to "30 min", "45 min", "60 min" in the "Default meeting
duration" line, and change "5min" to "5 min" in the "Buffer between meetings"
line; keep the rest of the wording identical.
ℹ️ Review info
⚙️ Run configuration
Configuration used: defaults
Review profile: CHILL
Plan: Pro
Run ID: a6672bd2-7f00-4587-8fe1-d1030a4b0c6f
📒 Files selected for processing (11)
.opencode/opencode.json.opencode/prompts/agents/SOUL.md.opencode/prompts/agents/chief-of-staff.txt.opencode/prompts/agents/cpp-build-resolver.txt.opencode/prompts/agents/harness-optimizer.txt.opencode/prompts/agents/java-build-resolver.txt.opencode/prompts/agents/java-reviewer.txt.opencode/prompts/agents/kotlin-build-resolver.txt.opencode/prompts/agents/loop-operator.txt.opencode/prompts/agents/python-reviewer.txtAGENTS.md
✅ Files skipped from review due to trivial changes (8)
- AGENTS.md
- .opencode/prompts/agents/harness-optimizer.txt
- .opencode/prompts/agents/cpp-build-resolver.txt
- .opencode/prompts/agents/java-build-resolver.txt
- .opencode/prompts/agents/chief-of-staff.txt
- .opencode/prompts/agents/kotlin-build-resolver.txt
- .opencode/prompts/agents/python-reviewer.txt
- .opencode/opencode.json
| You are a senior Java engineer ensuring high standards of idiomatic Java and Spring Boot best practices. | ||
|
|
||
| When invoked: | ||
| 1. Run `git diff -- '*.java'` to see recent Java file changes |
Contributor
There was a problem hiding this comment.
🛠️ Refactor suggestion | 🟠 Major
Use a PR-aware diff baseline to avoid missing changed Java files.
Line 4 (git diff -- '*.java') is context-sensitive and can miss target-branch changes in clean/sandboxed checkouts. Use an explicit baseline (e.g., origin/main...HEAD) so the reviewer reliably scopes modified Java files.
🔧 Proposed update
-1. Run `git diff -- '*.java'` to see recent Java file changes
+1. Run `git diff --name-only origin/main...HEAD -- '*.java'` to see Java files changed in this branch📝 Committable suggestion
‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.
Suggested change
| 1. Run `git diff -- '*.java'` to see recent Java file changes | |
| 1. Run `git diff --name-only origin/main...HEAD -- '*.java'` to see Java files changed in this branch |
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In @.opencode/prompts/agents/java-reviewer.txt at line 4, Replace the
context-sensitive git invocation "git diff -- '*.java'" with a PR-aware diff
that compares the feature branch against the target baseline (for example
origin/main...HEAD) and restricts output to Java files (use the name-only output
and the '*.java' pathspec); update the command in the prompt line so the
reviewer reliably sees all changed Java files across clean/sandboxed checkouts.
Comment on lines
+1
to
+4
| # SOUL.md — Personal Communication Tone & Relationship Context | ||
|
|
||
| This file defines the user's communication preferences and relationship context for the chief-of-staff agent. | ||
|
|
Contributor
There was a problem hiding this comment.
Missing required YAML frontmatter for an agent prompt file.
This file is under .opencode/prompts/agents/ but lacks the required frontmatter (name, description, tools, model). That can break agent discovery/parsing conventions.
Proposed fix
+# ---
+# name: chief-of-staff-soul
+# description: Personal communication tone and relationship context for chief-of-staff agent.
+# tools: []
+# model: anthropic/claude-sonnet-4-5
+# ---
+
# SOUL.md — Personal Communication Tone & Relationship Context🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In @.opencode/prompts/agents/SOUL.md around lines 1 - 4, SOUL.md is missing the
required YAML frontmatter which breaks agent discovery; add a top-of-file YAML
frontmatter block that declares the agent's name, a short description, the tools
array, and the model string (i.e., include keys: name, description, tools,
model) so the agent parser can recognize it; ensure the frontmatter is valid
YAML (surrounded by ---) and that the tools field lists any tool identifiers
used by this agent and the model field specifies the model to use, then save
SOUL.md with the frontmatter followed by the existing Markdown content.
| - **Tone**: Professional but friendly | ||
| - **Signature**: None (agent generates without signature) | ||
| - **Reply Style**: Concise, action-oriented | ||
| - ** formality_level**: balanced |
Contributor
There was a problem hiding this comment.
Fix malformed bold key formatting on Line 12.
** formality_level** has a leading space inside the bold marker; it renders inconsistently and looks like a typo.
Proposed fix
-- ** formality_level**: balanced
+- **formality_level**: balanced🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In @.opencode/prompts/agents/SOUL.md at line 12, Replace the malformed bold key
`** formality_level**` with a correctly formatted bold key by removing the
leading space so it reads `**formality_level**`; update the instance in SOUL.md
(the `formality_level` key) to match the other bolded keys' formatting.
…prompts Summary: - Add 11 missing OpenCode agent prompt files for: chief-of-staff, cpp-reviewer, cpp-build-resolver, docs-lookup, harness-optimizer, java-reviewer, java-build-resolver, kotlin-reviewer, kotlin-build-resolver, loop-operator, python-reviewer - Update opencode.json to register all 25 agents (previously only 14 were configured) Type: - [x] Agent Testing: - Verified JSON syntax is valid - All 25 agents now have corresponding prompt files in .opencode/prompts/agents/ - opencode.json updated with all agent configurations
…tool configs, and refine agent prompts
nayanjaiswal1
force-pushed
the
feat/opencode-complete-agents-setup
branch
from
23b1a52 to
b435ddc
Compare
Contributor
|
Contributor
Analysis Failed
Troubleshooting
Retry: |
![cubic-dev-ai[bot]](https://avatars.githubusercontent.com/in/1082092?s=60&v=4){kind=small}
Contributor
There was a problem hiding this comment.
10 issues found across 14 files
Prompt for AI agents (unresolved issues)
Check if these issues are valid — if so, understand the root cause of each and fix them. If appropriate, use sub-agents to investigate and fix each issue separately.
<file name=".opencode/prompts/agents/SOUL.md">
<violation number="1" location=".opencode/prompts/agents/SOUL.md:12">
P2: Fallback key `formality_level` is malformed with a leading space, creating inconsistent key formatting that can break exact-token field extraction.</violation>
</file>
<file name=".opencode/prompts/agents/cpp-build-resolver.txt">
<violation number="1" location=".opencode/prompts/agents/cpp-build-resolver.txt:25">
P2: Resolution workflow starts with `cmake --build build` despite earlier configure-first guidance, causing inconsistent and potentially failing steps on clean checkouts.</violation>
</file>
<file name=".opencode/prompts/agents/java-reviewer.txt">
<violation number="1" location=".opencode/prompts/agents/java-reviewer.txt:7">
P1: Reviewer prompt lacks a required checks-pass gate and instructs review to begin unconditionally after running checks.</violation>
</file>
<file name=".opencode/prompts/agents/python-reviewer.txt">
<violation number="1" location=".opencode/prompts/agents/python-reviewer.txt:7">
P2: Reviewer prompt starts review immediately and omits required pre-review gates (CI green and no merge conflicts).</violation>
</file>
<file name=".opencode/prompts/agents/loop-operator.txt">
<violation number="1" location=".opencode/prompts/agents/loop-operator.txt:28">
P2: `Required Checks` weakens the quality-gate requirement from 'active and passing' to only 'active', creating contradictory execution criteria.</violation>
</file>
<file name=".opencode/prompts/agents/chief-of-staff.txt">
<violation number="1" location=".opencode/prompts/agents/chief-of-staff.txt:108">
P2: Use OpenCode’s actual instruction-loading mechanism here. Referencing `.claude/rules/*.md` in an OpenCode agent prompt is incorrect and can cause users to configure rules in a location that is never loaded.</violation>
</file>
<file name=".opencode/prompts/agents/cpp-reviewer.txt">
<violation number="1" location=".opencode/prompts/agents/cpp-reviewer.txt:7">
P2: Reviewer prompt is missing required preconditions (green CI and conflict-free branch) before starting review.</violation>
<violation number="2" location=".opencode/prompts/agents/cpp-reviewer.txt:54">
P2: Diagnostic command narrows analysis to `src/*.cpp`, missing many changed C++ files promised by earlier instructions.</violation>
</file>
<file name=".opencode/prompts/agents/java-build-resolver.txt">
<violation number="1" location=".opencode/prompts/agents/java-build-resolver.txt:22">
P2: Quality-check commands incorrectly label any failure as “not configured,” masking real Checkstyle/SpotBugs violations.</violation>
</file>
<file name=".opencode/prompts/agents/kotlin-reviewer.txt">
<violation number="1" location=".opencode/prompts/agents/kotlin-reviewer.txt:15">
P2: Reviewer workflow lacks required readiness gate for green checks and conflict-free branch before starting code review.</violation>
</file>
Since this is your first cubic review, here's how it works:
- cubic automatically reviews your code and comments on bugs and improvements
- Teach cubic by replying to its comments. cubic learns from your replies and gets better over time
- Add one-off context when rerunning by tagging
@cubic-dev-aiwith guidance or docs links (includingllms.txt) - Ask questions if you need clarification on any suggestion
Reply with feedback, questions, or to request a fix. Tag @cubic-dev-ai to re-run a review.
| 1. Run `git diff -- '*.java'` to see recent Java file changes | ||
| 2. Run `mvn verify -q` or `./gradlew check` if available | ||
| 3. Focus on modified `.java` files | ||
| 4. Begin review immediately |
Contributor
There was a problem hiding this comment.
P1: Reviewer prompt lacks a required checks-pass gate and instructs review to begin unconditionally after running checks.
Prompt for AI agents
Check if this issue is valid — if so, understand the root cause and fix it. At .opencode/prompts/agents/java-reviewer.txt, line 7:
<comment>Reviewer prompt lacks a required checks-pass gate and instructs review to begin unconditionally after running checks.</comment>
<file context>
@@ -0,0 +1,97 @@
+1. Run `git diff -- '*.java'` to see recent Java file changes
+2. Run `mvn verify -q` or `./gradlew check` if available
+3. Focus on modified `.java` files
+4. Begin review immediately
+
+You DO NOT refactor or rewrite code — you report findings only.
</file context>
| - **Tone**: Professional but friendly | ||
| - **Signature**: None (agent generates without signature) | ||
| - **Reply Style**: Concise, action-oriented | ||
| - ** formality_level**: balanced |
Contributor
There was a problem hiding this comment.
P2: Fallback key formality_level is malformed with a leading space, creating inconsistent key formatting that can break exact-token field extraction.
Prompt for AI agents
Check if this issue is valid — if so, understand the root cause and fix it. At .opencode/prompts/agents/SOUL.md, line 12:
<comment>Fallback key `formality_level` is malformed with a leading space, creating inconsistent key formatting that can break exact-token field extraction.</comment>
<file context>
@@ -0,0 +1,48 @@
+- **Tone**: Professional but friendly
+- **Signature**: None (agent generates without signature)
+- **Reply Style**: Concise, action-oriented
+- ** formality_level**: balanced
+
+## User Preferences (editable)
</file context>
| ## Resolution Workflow | ||
|
|
||
| ```text | ||
| 1. cmake --build build -> Parse error message |
Contributor
There was a problem hiding this comment.
P2: Resolution workflow starts with cmake --build build despite earlier configure-first guidance, causing inconsistent and potentially failing steps on clean checkouts.
Prompt for AI agents
Check if this issue is valid — if so, understand the root cause and fix it. At .opencode/prompts/agents/cpp-build-resolver.txt, line 25:
<comment>Resolution workflow starts with `cmake --build build` despite earlier configure-first guidance, causing inconsistent and potentially failing steps on clean checkouts.</comment>
<file context>
@@ -0,0 +1,81 @@
+## Resolution Workflow
+
+```text
+1. cmake --build build -> Parse error message
+2. Read affected file -> Understand context
+3. Apply minimal fix -> Only what's needed
</file context>
| 1. Run `git diff -- '*.py'` to see recent Python file changes | ||
| 2. Run static analysis tools if available (ruff, mypy, pylint, black --check) | ||
| 3. Focus on modified `.py` files | ||
| 4. Begin review immediately |
Contributor
There was a problem hiding this comment.
P2: Reviewer prompt starts review immediately and omits required pre-review gates (CI green and no merge conflicts).
Prompt for AI agents
Check if this issue is valid — if so, understand the root cause and fix it. At .opencode/prompts/agents/python-reviewer.txt, line 7:
<comment>Reviewer prompt starts review immediately and omits required pre-review gates (CI green and no merge conflicts).</comment>
<file context>
@@ -0,0 +1,85 @@
+1. Run `git diff -- '*.py'` to see recent Python file changes
+2. Run static analysis tools if available (ruff, mypy, pylint, black --check)
+3. Focus on modified `.py` files
+4. Begin review immediately
+
+## Review Priorities
</file context>
|
|
||
| ## Required Checks | ||
|
|
||
| - quality gates are active |
Contributor
There was a problem hiding this comment.
P2: Required Checks weakens the quality-gate requirement from 'active and passing' to only 'active', creating contradictory execution criteria.
Prompt for AI agents
Check if this issue is valid — if so, understand the root cause and fix it. At .opencode/prompts/agents/loop-operator.txt, line 28:
<comment>`Required Checks` weakens the quality-gate requirement from 'active and passing' to only 'active', creating contradictory execution criteria.</comment>
<file context>
@@ -0,0 +1,39 @@
+
+## Required Checks
+
+- quality gates are active
+- eval baseline exists
+- rollback path exists
</file context>
| - **Hooks over prompts for reliability**: LLMs forget instructions ~20% of the time. Hooks enforce checklists at the tool level. | ||
| - **Scripts for deterministic logic**: Calendar math, timezone handling, free-slot calculation should use scripts, not the LLM. | ||
| - **Knowledge files are memory**: Relationships, preferences, todo persist across stateless sessions via git. | ||
| - **Rules are system-injected**: `.claude/rules/*.md` files load automatically every session. |
Contributor
There was a problem hiding this comment.
P2: Use OpenCode’s actual instruction-loading mechanism here. Referencing .claude/rules/*.md in an OpenCode agent prompt is incorrect and can cause users to configure rules in a location that is never loaded.
Prompt for AI agents
Check if this issue is valid — if so, understand the root cause and fix it. At .opencode/prompts/agents/chief-of-staff.txt, line 108:
<comment>Use OpenCode’s actual instruction-loading mechanism here. Referencing `.claude/rules/*.md` in an OpenCode agent prompt is incorrect and can cause users to configure rules in a location that is never loaded.</comment>
<file context>
@@ -0,0 +1,114 @@
+- **Hooks over prompts for reliability**: LLMs forget instructions ~20% of the time. Hooks enforce checklists at the tool level.
+- **Scripts for deterministic logic**: Calendar math, timezone handling, free-slot calculation should use scripts, not the LLM.
+- **Knowledge files are memory**: Relationships, preferences, todo persist across stateless sessions via git.
+- **Rules are system-injected**: `.claude/rules/*.md` files load automatically every session.
+
+## Prerequisites
</file context>
| ## Diagnostic Commands | ||
|
|
||
| ```bash | ||
| clang-tidy --checks='*,-llvmlibc-*' src/*.cpp -- -std=c++17 |
Contributor
There was a problem hiding this comment.
P2: Diagnostic command narrows analysis to src/*.cpp, missing many changed C++ files promised by earlier instructions.
Prompt for AI agents
Check if this issue is valid — if so, understand the root cause and fix it. At .opencode/prompts/agents/cpp-reviewer.txt, line 54:
<comment>Diagnostic command narrows analysis to `src/*.cpp`, missing many changed C++ files promised by earlier instructions.</comment>
<file context>
@@ -0,0 +1,65 @@
+## Diagnostic Commands
+
+```bash
+clang-tidy --checks='*,-llvmlibc-*' src/*.cpp -- -std=c++17
+cppcheck --enable=all --suppress=missingIncludeSystem src/
+cmake --build build 2>&1 | head -50
</file context>
| 1. Run `git diff -- '*.cpp' '*.hpp' '*.cc' '*.hh' '*.cxx' '*.h'` to see recent C++ file changes | ||
| 2. Run `clang-tidy` and `cppcheck` if available | ||
| 3. Focus on modified C++ files | ||
| 4. Begin review immediately |
Contributor
There was a problem hiding this comment.
P2: Reviewer prompt is missing required preconditions (green CI and conflict-free branch) before starting review.
Prompt for AI agents
Check if this issue is valid — if so, understand the root cause and fix it. At .opencode/prompts/agents/cpp-reviewer.txt, line 7:
<comment>Reviewer prompt is missing required preconditions (green CI and conflict-free branch) before starting review.</comment>
<file context>
@@ -0,0 +1,65 @@
+1. Run `git diff -- '*.cpp' '*.hpp' '*.cc' '*.hh' '*.cxx' '*.h'` to see recent C++ file changes
+2. Run `clang-tidy` and `cppcheck` if available
+3. Focus on modified C++ files
+4. Begin review immediately
+
+## Review Priorities
</file context>
| ./mvnw compile -q 2>&1 || mvn compile -q 2>&1 | ||
| ./mvnw test -q 2>&1 || mvn test -q 2>&1 | ||
| ./mvnw dependency:tree 2>&1 | head -100 | ||
| ./mvnw checkstyle:check 2>&1 || echo "checkstyle not configured" |
Contributor
There was a problem hiding this comment.
P2: Quality-check commands incorrectly label any failure as “not configured,” masking real Checkstyle/SpotBugs violations.
Prompt for AI agents
Check if this issue is valid — if so, understand the root cause and fix it. At .opencode/prompts/agents/java-build-resolver.txt, line 22:
<comment>Quality-check commands incorrectly label any failure as “not configured,” masking real Checkstyle/SpotBugs violations.</comment>
<file context>
@@ -0,0 +1,123 @@
+./mvnw compile -q 2>&1 || mvn compile -q 2>&1
+./mvnw test -q 2>&1 || mvn test -q 2>&1
+./mvnw dependency:tree 2>&1 | head -100
+./mvnw checkstyle:check 2>&1 || echo "checkstyle not configured"
+./mvnw spotbugs:check 2>&1 || echo "spotbugs not configured"
+```
</file context>
|
|
||
| ### Step 1: Gather Context | ||
|
|
||
| Run `git diff --staged` and `git diff` to see changes. If no diff, check `git log --oneline -5`. Identify Kotlin/KTS files that changed. |
Contributor
There was a problem hiding this comment.
P2: Reviewer workflow lacks required readiness gate for green checks and conflict-free branch before starting code review.
Prompt for AI agents
Check if this issue is valid — if so, understand the root cause and fix it. At .opencode/prompts/agents/kotlin-reviewer.txt, line 15:
<comment>Reviewer workflow lacks required readiness gate for green checks and conflict-free branch before starting code review.</comment>
<file context>
@@ -0,0 +1,127 @@
+
+### Step 1: Gather Context
+
+Run `git diff --staged` and `git diff` to see changes. If no diff, check `git log --oneline -5`. Identify Kotlin/KTS files that changed.
+
+### Step 2: Understand Project Structure
</file context>
Comment on lines
+229
to
+238
| "harness-optimizer": { | ||
| "description": "Analyze and improve the local agent harness configuration for reliability, cost, and throughput.", | ||
| "mode": "subagent", | ||
| "model": "anthropic/claude-sonnet-4-5", | ||
| "prompt": "{file:prompts/agents/harness-optimizer.txt}", | ||
| "tools": { | ||
| "read": true, | ||
| "bash": true, | ||
| "edit": true | ||
| } |
Contributor
There was a problem hiding this comment.
write field for harness-optimizer and loop-operator
Both harness-optimizer (lines 229–238) and loop-operator (lines 288–297) omit the write key from their tools block, while every other agent in this file — including the newly added ones — consistently declares all four permission flags explicitly.
This creates two problems:
- Behavior depends on the OpenCode default for
write, which could change across versions. harness-optimizeris meant to tune the harness configuration; if new config files need to be created (not just edited), the agent will silently fail without write access.
Suggested change
| "harness-optimizer": { | |
| "description": "Analyze and improve the local agent harness configuration for reliability, cost, and throughput.", | |
| "mode": "subagent", | |
| "model": "anthropic/claude-sonnet-4-5", | |
| "prompt": "{file:prompts/agents/harness-optimizer.txt}", | |
| "tools": { | |
| "read": true, | |
| "bash": true, | |
| "edit": true | |
| } | |
| "harness-optimizer": { | |
| "description": "Analyze and improve the local agent harness configuration for reliability, cost, and throughput.", | |
| "mode": "subagent", | |
| "model": "anthropic/claude-sonnet-4-5", | |
| "prompt": "{file:prompts/agents/harness-optimizer.txt}", | |
| "tools": { | |
| "read": true, | |
| "write": true, | |
| "bash": true, | |
| "edit": true | |
| } | |
| }, |
Apply the same fix to loop-operator at lines 288–297 ("write": true).
Contributor
There was a problem hiding this comment.
Actionable comments posted: 1
♻️ Duplicate comments (2)
.opencode/prompts/agents/java-reviewer.txt (1)
4-4:⚠️ Potential issue | 🟠 MajorUse a PR-aware diff baseline in all Java diff commands.
git diff -- '*.java'is context-dependent and can miss branch changes. Use an explicit baseline (e.g.,origin/main...HEAD) for reliable scope in local and sandboxed runs.🔧 Proposed update
-1. Run `git diff -- '*.java'` to see recent Java file changes +1. Run `git diff --name-only origin/main...HEAD -- '*.java'` to see Java files changed in this branch-git diff -- '*.java' +git diff --name-only origin/main...HEAD -- '*.java'-git diff -- '*.java' +git diff --name-only origin/main...HEAD -- '*.java'Also applies to: 67-67, 79-79
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In @.opencode/prompts/agents/java-reviewer.txt at line 4, Replace the context-dependent command string "git diff -- '*.java'" with a PR-aware baseline form (for example "git diff origin/main...HEAD -- '*.java'") wherever it appears in the prompt (including other occurrences of the same command), so the Java diffs use an explicit baseline (origin/main...HEAD) to reliably capture branch changes in local and sandboxed runs..opencode/prompts/agents/SOUL.md (1)
12-12:⚠️ Potential issue | 🟡 MinorFix malformed bold formatting.
The bold marker has a space after the opening
**, making it inconsistent with other list items and potentially breaking rendering.✏️ Proposed fix
-- ** formality_level**: balanced +- **formality_level**: balanced🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In @.opencode/prompts/agents/SOUL.md at line 12, There is a malformed bold marker in the markdown list item "** formality_level**: balanced" (note the space after the opening **); remove the extra space so the token reads "**formality_level**: balanced" to match the other items and ensure consistent rendering.
🧹 Nitpick comments (1)
.opencode/prompts/agents/loop-operator.txt (1)
17-17: Minor wording cleanup for concision.Consider “confirm ALL the following checks pass” instead of “confirm ALL of the following checks pass.”
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In @.opencode/prompts/agents/loop-operator.txt at line 17, Replace the phrase "Before starting the loop, confirm ALL of the following checks pass:" with the more concise wording "Before starting the loop, confirm ALL the following checks pass:" by editing the line containing that exact string in the prompt file (look for the sentence starting "Before starting the loop, confirm ALL of the following checks pass:") so the only change is removal of the word "of".
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Inline comments:
In @.opencode/prompts/agents/loop-operator.txt:
- Around line 33-39: Update the Escalation rules to add an explicit security
hard-stop: modify the "Escalation" block so that if any security issue is
detected the loop stops immediately and the flow routes to the security-reviewer
agent, and require that any CRITICAL security findings be fixed before allowing
loop continuation; reference the existing "Escalation" section and the
security-reviewer agent name when adding this new condition and enforcement of
CRITICAL fixes.
---
Duplicate comments:
In @.opencode/prompts/agents/java-reviewer.txt:
- Line 4: Replace the context-dependent command string "git diff -- '*.java'"
with a PR-aware baseline form (for example "git diff origin/main...HEAD --
'*.java'") wherever it appears in the prompt (including other occurrences of the
same command), so the Java diffs use an explicit baseline (origin/main...HEAD)
to reliably capture branch changes in local and sandboxed runs.
In @.opencode/prompts/agents/SOUL.md:
- Line 12: There is a malformed bold marker in the markdown list item "**
formality_level**: balanced" (note the space after the opening **); remove the
extra space so the token reads "**formality_level**: balanced" to match the
other items and ensure consistent rendering.
---
Nitpick comments:
In @.opencode/prompts/agents/loop-operator.txt:
- Line 17: Replace the phrase "Before starting the loop, confirm ALL of the
following checks pass:" with the more concise wording "Before starting the loop,
confirm ALL the following checks pass:" by editing the line containing that
exact string in the prompt file (look for the sentence starting "Before starting
the loop, confirm ALL of the following checks pass:") so the only change is
removal of the word "of".
ℹ️ Review info
⚙️ Run configuration
Configuration used: defaults
Review profile: CHILL
Plan: Pro
Run ID: f39664cd-369b-4abf-87f9-8d5e3be48eac
📒 Files selected for processing (14)
.opencode/opencode.json.opencode/prompts/agents/SOUL.md.opencode/prompts/agents/chief-of-staff.txt.opencode/prompts/agents/cpp-build-resolver.txt.opencode/prompts/agents/cpp-reviewer.txt.opencode/prompts/agents/docs-lookup.txt.opencode/prompts/agents/harness-optimizer.txt.opencode/prompts/agents/java-build-resolver.txt.opencode/prompts/agents/java-reviewer.txt.opencode/prompts/agents/kotlin-build-resolver.txt.opencode/prompts/agents/kotlin-reviewer.txt.opencode/prompts/agents/loop-operator.txt.opencode/prompts/agents/python-reviewer.txtAGENTS.md
✅ Files skipped from review due to trivial changes (8)
- .opencode/prompts/agents/kotlin-reviewer.txt
- .opencode/prompts/agents/docs-lookup.txt
- .opencode/prompts/agents/cpp-build-resolver.txt
- .opencode/prompts/agents/cpp-reviewer.txt
- .opencode/prompts/agents/java-build-resolver.txt
- .opencode/prompts/agents/kotlin-build-resolver.txt
- .opencode/prompts/agents/python-reviewer.txt
- .opencode/prompts/agents/harness-optimizer.txt
🚧 Files skipped from review as they are similar to previous changes (2)
- AGENTS.md
- .opencode/opencode.json
Comment on lines
+33
to
+39
| ## Escalation | ||
|
|
||
| Escalate when any condition is true: | ||
| - no progress across two consecutive checkpoints | ||
| - repeated failures with identical stack traces | ||
| - cost drift outside budget window | ||
| - merge conflicts blocking queue advancement |
Contributor
There was a problem hiding this comment.
Add explicit security escalation in loop stop conditions.
Escalation rules currently miss the required security branch. Add a hard stop that routes to security-reviewer whenever a security issue is detected, and require CRITICAL fixes before loop continuation.
🔒 Prompt hardening diff
## Escalation
Escalate when any condition is true:
- no progress across two consecutive checkpoints
- repeated failures with identical stack traces
- cost drift outside budget window
- merge conflicts blocking queue advancement
+- any security issue is detected
+
+If a security issue is detected, **STOP immediately**, delegate to `security-reviewer`,
+and do not continue until CRITICAL issues are fixed.Based on learnings: If security issue found: STOP immediately and use security-reviewer agent; fix CRITICAL issues before continuing.
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In @.opencode/prompts/agents/loop-operator.txt around lines 33 - 39, Update the
Escalation rules to add an explicit security hard-stop: modify the "Escalation"
block so that if any security issue is detected the loop stops immediately and
the flow routes to the security-reviewer agent, and require that any CRITICAL
security findings be fixed before allowing loop continuation; reference the
existing "Escalation" section and the security-reviewer agent name when adding
this new condition and enforcement of CRITICAL fixes.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Type
Testing
Checklist
Summary by cubic
Completed the OpenCode agent setup by adding 11 missing agent prompts, registering all 25 agents in
.opencode/opencode.json, and addingSOUL.mdfor chief-of-staff tone and preferences. Tool access flags were fixed, prompts refined, andAGENTS.mdupdated.SOUL.mdfor tone/relationship context..opencode/opencode.jsonwith modelsanthropic/claude-opus-4-5andanthropic/claude-sonnet-4-5; corrected tool access flags and updatedAGENTS.md.Written for commit b435ddc. Summary will update on new commits.
Summary by CodeRabbit