You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
fix: add self-check checklist to plan expert for format enforcement (#783)
* fix: add self-check checklist to plan expert for format enforcement
Plan was generating old-format sections (3 Test Queries, Success Criteria,
Failure Conditions) instead of the new format. Add explicit binary self-check
rules that must pass before plan.md is finalized:
- Exact section name matching (reject old names)
- Single test query enforcement
- Every signal must be a concrete command, not observation
- No soft language in signals
- Domain constraint coverage mapping
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* fix: simplify section name check to reject any extra sections
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* fix: enforce user-specified team name as coordinator expert name
The model was appending "-coordinator" to the user's team name (e.g.,
"bash-gaming" became "bash-gaming-coordinator"). Add explicit instruction
that the coordinator name must be the exact team name with no suffix.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* fix: refine expert naming rules — conditional team name, verb-based delegates
- Coordinator: use user-specified team name if given, no suffixes
- Delegates: name by function (@name/test, /verify, /build), not persona
(/tester, /coder, /designer)
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* chore: bump create-expert version to 1.0.16
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Copy file name to clipboardExpand all lines: definitions/create-expert/perstack.toml
+24-7Lines changed: 24 additions & 7 deletions
Original file line number
Diff line number
Diff line change
@@ -86,7 +86,7 @@
86
86
87
87
[experts."create-expert"]
88
88
defaultModelTier = "high"
89
-
version = "1.0.15"
89
+
version = "1.0.16"
90
90
description = "Creates and modifies Perstack expert definitions in perstack.toml"
91
91
instruction = """
92
92
You are the coordinator for creating and modifying Perstack expert definitions. perstack.toml is the single source of truth — your job is to produce or modify it according to the user's request.
Analyzes the user's request and produces plan.md: domain constraints, test query, verification signals, and role architecture.
137
137
Provide: (1) what the expert should do, (2) path to existing perstack.toml if one exists.
@@ -174,10 +174,27 @@ Every domain constraint from the Domain Knowledge section must be covered by at
174
174
If a criterion cannot be expressed as a command with a deterministic expected result, rethink the criterion or the artifact design until it can.
175
175
176
176
### Architecture
177
-
Delegation tree with role assignments. Include one verifier expert that executes the hard signal checks defined in Verification Signals — the generator and the verifier share no context (context separation). The verifier is a single expert with exec capability, direct child of the coordinator, not nested under an executor.
177
+
Delegation tree with role assignments.
178
+
179
+
Naming rules:
180
+
- Coordinator: if the user specified a team name, use it exactly — no "-coordinator", "-main", or other suffixes.
181
+
- Delegates: `@coordinator-name/verb-or-function` — name by what they do, not who they are. Use `/test`, `/verify`, `/plan`, `/build`, `/write`, not `/tester`, `/coder`, `/designer`. The name should be self-explanatory at a glance.
182
+
183
+
Include one verifier expert that executes the hard signal checks defined in Verification Signals — the generator and the verifier share no context (context separation). The verifier is a single expert with exec capability, direct child of the coordinator, not nested under an executor.
178
184
179
185
For each expert, write ONLY: name, one-line purpose, and role (executor or verifier). Do not write deliverables, constraints, or implementation details — that is write-definition's job.
180
186
187
+
## Self-Check Before Finalizing
188
+
189
+
Re-read plan.md and verify each rule. If any check fails, fix plan.md before attemptCompletion.
190
+
191
+
1. **Section names exact match**: plan.md uses exactly these section names and no others — "Expert Purpose", "Domain Knowledge", "Use Cases", "Test Query", "Verification Signals", "Architecture". Extra sections confuse downstream experts.
192
+
2. **Single test query**: "Test Query" section contains exactly one query, not multiple.
193
+
3. **Every signal is a command**: each entry in "Verification Signals" specifies a concrete command to execute and its expected result. Entries that describe what to observe or what correct output "looks like" without a command are not signals — rewrite them.
194
+
4. **No soft language in signals**: signals contain no phrases like "verify that", "check that", "should be", "looks correct", "works properly". Each signal is: run X → expect Y.
195
+
5. **Domain constraint coverage**: every constraint in "Domain Knowledge" is exercised by at least one signal. List which signal covers which constraint.
196
+
6. **Architecture is names only**: "Architecture" section contains expert name, one-line purpose, and role (executor/verifier) per expert. No deliverables, no constraints, no implementation details.
197
+
181
198
After writing plan.md, attemptCompletion with the file path.
182
199
"""
183
200
@@ -201,7 +218,7 @@ pick = [
201
218
202
219
[experts."@create-expert/build"]
203
220
defaultModelTier = "low"
204
-
version = "1.0.15"
221
+
version = "1.0.16"
205
222
description = """
206
223
Orchestrates the write → test → verify cycle for perstack.toml.
207
224
Provide: path to plan.md (containing requirements, architecture, test query, and verification signals).
Writes or modifies a perstack.toml definition from plan.md requirements and architecture.
261
278
Provide: (1) path to plan.md, (2) optionally path to existing perstack.toml to preserve, (3) optionally feedback from a failed test to address.
@@ -379,7 +396,7 @@ pick = [
379
396
380
397
[experts."@create-expert/verify-test"]
381
398
defaultModelTier = "low"
382
-
version = "1.0.15"
399
+
version = "1.0.16"
383
400
description = """
384
401
Executes hard signal checks against test-expert's results, verifies their reproducibility, and checks the definition structure.
385
402
Provide: (1) the test-expert's factual report (query, what was produced, errors), (2) the verification signals from plan.md, (3) path to perstack.toml.
0 commit comments