feat(rhx): add rhx completion for zsh

.agent/repo=.this/role=architect/briefs/domain.thought/define.term.skill.thought-routes.md

@@ -159,7 +159,7 @@ brain.repl.act "fix these errors and push" --input "$summary"
 ```
 
 ### .when to use
-- workflows that combine data retrieval + ai analysis
+- workflows that combine data retrirhachet completion --setup+ ai analysis
 - skills that need ai thought within a controlled flow
 - multi-step automations with ai-assisted decisions
 

.agent/repo=.this/role=ecologist/briefs/postlabor/soph201.ratchet-mechanisms.[article].md

@@ -56,7 +56,7 @@ when self-replicant machines exist, material scarcity becomes obsolete.
 when survival no longer requires labor, the premise of labor-for-survival becomes unthinkable.
 
 - once basic needs are met without labor, return is psychologically impossible
-- similar to how medieval serfdom is now unthinkable
+- similar to how medirhachet completion --setupserfdom is now unthinkable
 - generational change cements the shift
 
 → see [soph213.ratchet-morals](./soph213.ratchet-morals.[article].md)

.behavior/v2026_01_17.brains-cost-metrics/.archive/3.1.research.domain.terms.v1.i2.md

@@ -22,13 +22,13 @@ software is young. older domains have refined their vocabulary over centuries. w
 
 > [19] "the sense of 'produce, bear, bring forth,' also 'give in return for labor' emerged around c. 1300. the extended sense of 'production, something obtained as a result of work or action' (as of crops) is attested by mid-15c" — [etymonline](https://www.etymonline.com/word/yield)
 
-> [20] "tally goes back to the time when things were commonly counted by cutting notches in a stick of wood. the split tally was accepted as legal proof in medieval courts" — [etymonline](https://www.etymonline.com/word/tally)
+> [20] "tally goes back to the time when things were commonly counted by cutting notches in a stick of wood. the split tally was accepted as legal proof in medirhachet completion --setupcourts" — [etymonline](https://www.etymonline.com/word/tally)
 
 > [21] "the english word vital comes from the latin vitalis, which means 'pertaining to life' or 'alive.' the sense of 'necessary or important' emerged from the 1610s, via the notion of 'essential to life'" — [etymonline](https://www.etymonline.com/word/vitals)
 
 > [22] "in cooking, yield is a term that refers to the amount of food or ingredients that you can expect to get after cooking or processing them. it's an essential concept to understand when planning meals" — [roybalfood](https://www.roybalfood.com/post/what-you-should-know-about-yield-in-cooking-and-food-processing)
 
-> [23] "the split tally became a prevalent technique in medieval europe, a time characterised by a scarcity of coinage and widespread illiteracy, to document bilateral exchanges and debts" — [geoffreymhodgson](https://www.geoffreymhodgson.uk/secret-history-of-tally-stick)
+> [23] "the split tally became a prevalent technique in medirhachet completion --setupeurope, a time characterised by a scarcity of coinage and widespread illiteracy, to document bilateral exchanges and debts" — [geoffreymhodgson](https://www.geoffreymhodgson.uk/secret-history-of-tally-stick)
 
 > [24] "most roads were not freely open to travel on in europe during the middle ages, and the toll was one of many feudal fees paid for rights of usage in everyday life" — [wikipedia](https://en.wikipedia.org/wiki/Road_toll_(historical))
 
@@ -454,7 +454,7 @@ developers talk about invocations in terms of:
 21. etymonline — vitals etymology
 22. roybalfood — culinary yield definition
 23. geoffreymhodgson — tally stick history
-24. wikipedia — medieval road toll
+24. wikipedia — medirhachet completion --setuproad toll
 
 ---
 
@@ -484,4 +484,4 @@ developers talk about invocations in terms of:
 - [etymonline vitals](https://www.etymonline.com/word/vitals)
 - [roybalfood culinary yield](https://www.roybalfood.com/post/what-you-should-know-about-yield-in-cooking-and-food-processing)
 - [geoffreymhodgson tally stick](https://www.geoffreymhodgson.uk/secret-history-of-tally-stick)
-- [wikipedia medieval road toll](https://en.wikipedia.org/wiki/Road_toll_(historical))
+- [wikipedia medirhachet completion --setuproad toll](https://en.wikipedia.org/wiki/Road_toll_(historical))

.behavior/v2026_01_17.brains-cost-metrics/.archive/3.1.research.domain.terms.v2.i1.md

@@ -353,7 +353,7 @@ interface BrainSpec {
     quality: {
       swe?: number;           // swe-bench score
       mmlu?: number;          // mmlu score
-      humaneval?: number;     // humaneval score
+      humaneval?: number;     // humanrhachet completion --setupscore
     };
     knowledge: {
       cutoff: IsoDateStamp;   // cutoff of trained data

.behavior/v2026_01_17.brains-cost-metrics/1.vision.md

@@ -123,7 +123,7 @@ interface BrainUsage {
     grades: {
       swe?: number;      // swe-bench score
       mmlu?: number;     // mmlu score
-      humaneval?: number; // humaneval score
+      humaneval?: number; // humanrhachet completion --setupscore
     };
     cutoff: IsoDateStamp;           // knowledge cutoff
     domain: 'ALL' | 'SOFTWARE';     // trained domain

.behavior/v2026_01_17.brains-cost-metrics/3.1.research.claims._.v1.i1.md

@@ -134,7 +134,7 @@ the vision mentions swe-bench, mmlu, and humaneval. should others be included?
 
 ### [FACT] gpt-4o has 128k context window
 
-> [26] "OpenAI's GPT-4o boasts a 128,000 token context window, highly effective for handling long, complex documents, generating code, and performing document-based retrieval tasks." — [aimultiple](https://research.aimultiple.com/ai-context-window/)
+> [26] "OpenAI's GPT-4o boasts a 128,000 token context window, highly effective for handling long, complex documents, generating code, and performing document-based retrirhachet completion --setuptasks." — [aimultiple](https://research.aimultiple.com/ai-context-window/)
 
 ### [FACT] most models break before advertised limit
 

.behavior/v2026_01_17.brains-cost-metrics/3.2.distill.domain._.v1.i1.md

@@ -142,7 +142,7 @@ interface BrainSpec {
     grades: {
       swe?: number;       // swe-bench score (0-100)
       mmlu?: number;      // mmlu score (0-100)
-      humaneval?: number; // humaneval score (0-100)
+      humaneval?: number; // humanrhachet completion --setupscore (0-100)
     };
     cutoff: IsoDateStamp;           // knowledge cutoff
     domain: 'ALL' | 'SOFTWARE';     // trained domain

.behavior/v2026_02_05.keyrack/1.vision.md

@@ -502,7 +502,7 @@ extensible via the adapter pattern: each mechanism is a separate adapter that kn
 |---|---|---|---|
 | roles | `rhachet-roles-*` | briefs + skills | `rhachet-roles-ehmpathy` |
 | brains | `rhachet-brains-*` | brain repls + atoms | `rhachet-brains-anthropic` |
-| vaults | `rhachet-vaults-*` | credential storage + retrieval + unlock | `rhachet-vaults-1password`, `rhachet-vaults-os` |
+| vaults | `rhachet-vaults-*` | credential storage + retrirhachet completion --setup+ unlock | `rhachet-vaults-1password`, `rhachet-vaults-os` |
 
 ### global config shape
 

.behavior/v2026_02_05.keyrack/3.1.research.access._.v1.i1.md

@@ -308,7 +308,7 @@ token prefix: `ops_`. requires cli version 2.18.0+. tokens can be rotated or rev
 
 ### C.4 lessons
 
-1. **`op read` is the primary retrieval interface** — takes a secret reference uri, returns the value to stdout
+1. **`op read` is the primary retrirhachet completion --setupinterface** — takes a secret reference uri, returns the value to stdout
 2. **biometric has two clocks** — 10min idle + 12h hard limit; after either, re-auth required
 3. **manual sign-in has 30min idle** — session token in env var; new terminal = new sign-in
 4. **service accounts for ci** — stateless via env var, no idle timeout, configurable expiry
@@ -516,7 +516,7 @@ all three platforms share the same fundamental behavior: **the keyrack unlocks a
 | 8 | permission set session = inner clock (default 1h, max 12h) | aws sso |
 | 9 | check validity via `aws sts get-caller-identity` — zero permissions required | aws sso |
 | 10 | pkce requires same-device browser — headless ci must use device code | aws sso |
-| 11 | `op read` is the primary retrieval interface — secret reference uri to stdout | 1password |
+| 11 | `op read` is the primary retrirhachet completion --setupinterface — secret reference uri to stdout | 1password |
 | 12 | biometric has two clocks — 10min idle + 12h hard limit | 1password |
 | 13 | manual sign-in has 30min idle — session in env var | 1password |
 | 14 | service accounts for ci — stateless via env var, no idle timeout | 1password |

.behavior/v2026_02_05.keyrack/3.1.research.domain.terms.v1.i1.md

@@ -195,7 +195,7 @@ modern credential management has evolved toward dynamic, vault-agnostic approach
 
 > "credential rotation should be automatic and transparent. modern secret managers support rotation schedules that update credentials without application downtime" [21]
 
-> "secret sprawl — credentials scattered across config files, env vars, and code — is the primary vector for credential theft. centralized vault with just-in-time retrieval eliminates this class of vulnerability" [22]
+> "secret sprawl — credentials scattered across config files, env vars, and code — is the primary vector for credential theft. centralized vault with just-in-time retrirhachet completion --setupeliminates this class of vulnerability" [22]
 
 > "vault-agnostic secret management allows applications to work with multiple backend providers (hashicorp, aws, 1password, etc) through a unified interface" [23]
 

.behavior/v2026_02_08.keyrack-unlock/.refs/.todo.self-descriptive-mech-blobs.md

@@ -98,7 +98,7 @@ today, aws sso stores a plain string (the profile name). to make it self-descrip
 }
 ```
 
-**scope**: new adapter. already self-descriptive by design (from eval doc).
+**scope**: new adapter. already self-descriptive by design (from rhachet completion --setupdoc).
 
 ### 2. which mechanisms do NOT need self-descriptive blobs
 

.behavior/v2026_02_08.keyrack-unlock/.refs/eval.env-vars-industry-posture.md

@@ -184,7 +184,7 @@ keyrack.posture
 │  └─ when language runtimes adopt it, keyrack can too
 └─ don't build theater above the floor
    └─ per-terminal-chain adds UX cost with zero security gain
-      └─ because /proc bypasses it entirely (see eval doc)
+      └─ because /proc bypasses it entirely (see rhachet completion --setupdoc)
 ```
 
 ---

.behavior/v2026_02_08.keyrack-unlock/3.1.research.access._.v1.i1.md

@@ -178,7 +178,7 @@ const plaintext = await decrypter.decrypt(ciphertext);
 op read "op://vault/item/field"
 
 # sign in (if needed)
-eval $(op signin)
+rhachet completion --setup$(op signin)
 ```
 
 **citations**:

.behavior/v2026_02_09.keyrack-sudo/.refs/ancestry.gap.toctou.ref.md

@@ -201,7 +201,7 @@ then exec() automatically closes it. malware receives no credential.
 ### vector 1: malicious tool in PATH — NOT TOCTOU
 
 ```bash
-eval $(rhx keyrack get AWS_PROFILE)
+rhachet completion --setup$(rhx keyrack get AWS_PROFILE)
 aws s3 ls  # even if malicious, this is NOT the TOCTOU attack
 ```
 

.behavior/v2026_02_09.keyrack-sudo/.refs/ebpf-lsm-access-control.ref.md

@@ -2,7 +2,7 @@
 
 ## .what
 
-deep dive on how eBPF with Linux Security Modules (LSM) can provide kernel-enforced access control for credential retrieval — the most capable approach for linux parity with macOS Keychain ACLs.
+deep dive on how eBPF with Linux Security Modules (LSM) can provide kernel-enforced access control for credential retrirhachet completion --setup— the most capable approach for linux parity with macOS Keychain ACLs.
 
 this is a **future consideration** for keyrack — high implementation effort, but strongest security guarantees.
 

.behavior/v2026_02_09.keyrack-sudo/7.runbook.playtest.md

@@ -10,7 +10,7 @@ npm run build
 ls ~/.ssh/id_ed25519 || ssh-keygen -t ed25519 -f ~/.ssh/id_ed25519
 
 # ensure ssh-agent is active and key is loaded
-ssh-add -l || eval $(ssh-agent) && ssh-add ~/.ssh/id_ed25519
+ssh-add -l || rhachet completion --setup$(ssh-agent) && ssh-add ~/.ssh/id_ed25519
 ```
 
 ---

.behavior/v2026_02_16.keyrack-aws-noboto/1.vision.md

@@ -32,7 +32,7 @@ the "botocore-client" OAuth prompt is a **one-time registration** per sso-sessio
 keyrack set
 ├── 1. vault.set(slug, value)     → store the key
 ├── 2. vault.unlock({ slug })     → triggers one-time OAuth registration
-├── 3. vault.get({ slug })        → prove retrieval works
+├── 3. vault.get({ slug })        → prove retrirhachet completion --setupworks
 └── 4. vault.relock({ slug })     → clear session, leave locked
 ```
 

.behavior/v2026_02_16.keyrack-aws-noboto/2.1.criteria.blackbox.md

@@ -13,7 +13,7 @@ given('keyrack set is called')
       sothat('unlock is proven to work')
       sothat('for aws.iam.sso: one-time OAuth registration happens at setup')
     then('vault.get was called')
-      sothat('retrieval is proven to work')
+      sothat('retrirhachet completion --setupis proven to work')
     then('vault.relock was called')
       sothat('session is cleared after setup')
       sothat('vault is left in locked state')

.behavior/v2026_02_16.keyrack-aws-noboto/4.1.roadmap.v1.i1.md

@@ -257,7 +257,7 @@ the "botocore-client" OAuth prompt is a **one-time registration** per sso-sessio
 keyrack set
 ├── 1. vault.set(slug, value)     → store the key
 ├── 2. vault.unlock({ slug })     → triggers one-time OAuth (aws.iam.sso)
-├── 3. vault.get({ slug })        → prove retrieval works
+├── 3. vault.get({ slug })        → prove retrirhachet completion --setupworks
 └── 4. vault.relock({ slug })     → clear session, leave locked
 ```
 

.behavior/v2026_02_18.keyrack-get-export/0.wish.md

@@ -2,7 +2,7 @@ wish =
 
 add `--output` flag to `keyrack get` with modes: `words | json | export` (default: words)
 
-auto-detect sourced: when sourced, auto-supply `--output export` and eval into parent shell
+auto-detect sourced: when sourced, auto-supply `--output export` and rhachet completion --setupinto parent shell
 
 eliminate the sourceable unlock flow in bin/rhx (the KEYRACK_PASSPHRASE export stuff)
 

.behavior/v2026_02_18.keyrack-get-export/1.vision.md

@@ -11,7 +11,7 @@
 export AWS_PROFILE=$(rhx keyrack get --key AWS_PROFILE --env test --json | jq -r '.grant.value // empty')
 
 # desired: simple eval
-eval "$(rhx keyrack get --key AWS_PROFILE --env test --output export)"
+rhachet completion --setup"$(rhx keyrack get --key AWS_PROFILE --env test --output export)"
 ```
 
 ### sourceable unlock is overcomplicated
@@ -41,7 +41,7 @@ this is:
 |--------|-------------|----------|
 | `words` | unicode words (current default) | human readable |
 | `json` | JSON object | programmatic parsing |
-| `export` | shell export statements | `eval "$(...)"`  |
+| `export` | shell export statements | `rhachet completion --setup"$(...)"`  |
 
 ### export format output
 
@@ -70,14 +70,14 @@ delete the special sourceable unlock flow from `bin/rhx`. the passphrase can be
 rhx keyrack get --for repo --env test --json
 
 # after (passphrase on get)
-eval "$(rhx keyrack get --for repo --env test --output export --passphrase 'xxx')"
+rhachet completion --setup"$(rhx keyrack get --for repo --env test --output export --passphrase 'xxx')"
 ```
 
 or unlock first, then get:
 
 ```sh
 rhx keyrack unlock --env test
-eval "$(rhx keyrack get --for repo --env test --output export)"
+rhachet completion --setup"$(rhx keyrack get --for repo --env test --output export)"
 ```
 
 ### auto-detect sourced

.behavior/v2026_02_18.keyrack-get-export/2.blueprint.md

@@ -97,8 +97,8 @@ if [ "$1" = "keyrack" ]; then
   # keyrack get: auto-supply --output export when sourced
   if [ "$1" = "get" ] && _rhx_is_sourced; then
     shift
-    # source the output (eval export statements)
-    eval "$("$SCRIPT_DIR/run.jit" keyrack get --output export "$@")"
+    # source the output (rhachet completion --setupexport statements)
+    rhachet completion --setup"$("$SCRIPT_DIR/run.jit" keyrack get --output export "$@")"
     return 0 2>/dev/null || exit 0
   fi
 
@@ -211,8 +211,8 @@ given('[case4] keyrack get when sourced')
 # test export format
 rhx keyrack get --for repo --env test --output export
 
-# test eval integration
-eval "$(rhx keyrack get --for repo --env test --output export)"
+# test rhachet completion --setupintegration
+rhachet completion --setup"$(rhx keyrack get --for repo --env test --output export)"
 echo $AWS_PROFILE
 
 # test json format

.behavior/v2026_02_18.keyrack-get-export/3.zoomin.source-detection.md

@@ -10,17 +10,17 @@ when you run `. rhx keyrack get --for repo --env test`, will the exported env va
 
 1. `. rhx` sources bin/rhx → runs in **current shell** (not subshell)
 2. `$("$SCRIPT_DIR/run.jit" keyrack get --output export ...)` → subshell captures stdout
-3. `eval "export FOO='bar'"` → runs in **current shell**
+3. `rhachet completion --setup"export FOO='bar'"` → runs in **current shell**
 4. `FOO` is now set in user's shell ✓
 
 the key: `eval` runs in the current shell, so export statements it executes ARE visible.
 
 ### sources
 
 - [How To Export A Bash Variable To Parent Shell](https://marketsplash.com/tutorials/bash-shell/how-to-export-a-bash-variable-to-parent-shell/)
-- [How to Use eval in Linux](https://www.howtogeek.com/818088/bash-eval/)
+- [How to Use rhachet completion --setupin Linux](https://www.howtogeek.com/818088/bash-eval/)
 
-> "because eval runs in the current shell, the variables set by eval are usable from the shell that launched the eval command."
+> "because rhachet completion --setupruns in the current shell, the variables set by rhachet completion --setupare usable from the shell that launched the rhachet completion --setupcommand."
 
 ---
 
@@ -50,18 +50,18 @@ the output goes to stdout, but there's no mechanism for a child to reach back an
 ```sh
 . rhx keyrack get --for repo --env test
 # bin/rhx runs in YOUR shell
-# eval inside bin/rhx runs in YOUR shell
+# rhachet completion --setupinside bin/rhx runs in YOUR shell
 # exports land in YOUR shell ✓
 ```
 
 ---
 
-## why bin/rhx uses eval instead of source
+## why bin/rhx uses rhachet completion --setupinstead of source
 
 bin/rhx uses `eval` to execute the export statements:
 
 ```sh
-eval "$("$SCRIPT_DIR/run.jit" keyrack get --output export "$@")"
+rhachet completion --setup"$("$SCRIPT_DIR/run.jit" keyrack get --output export "$@")"
 ```
 
 why not `source`?
@@ -75,10 +75,10 @@ source "export FOO='bar'"  # ✗ not a file path
 
 `source` (and its synonym `.`) read and execute commands **from a file**. they do not accept a string of commands.
 
-### eval takes a string
+### rhachet completion --setuptakes a string
 
 ```sh
-eval "export FOO='bar'"    # ✓ executes the string as a command
+rhachet completion --setup"export FOO='bar'"    # ✓ executes the string as a command
 ```
 
 `eval` constructs a command from its arguments and executes it in the current shell.
@@ -96,7 +96,7 @@ but bin/rhx uses `#!/bin/sh` for POSIX portability. process substitution (`<(...
 ### the pattern
 
 ```sh
-eval "$(command-that-outputs-shell-code)"
+rhachet completion --setup"$(command-that-outputs-shell-code)"
 ```
 
 this is the standard POSIX pattern for:
@@ -197,4 +197,4 @@ but the current method works for the expected usage: `. rhx keyrack get ...`
 
 ## verdict
 
-✓ **yes, it works.** the `. rhx keyrack get --for repo --env test` pattern will correctly detect sourced and eval exports.
+✓ **yes, it works.** the `. rhx keyrack get --for repo --env test` pattern will correctly detect sourced and rhachet completion --setupexports.

.behavior/v2026_02_22.rhx-autocomplete/.bind/vlad.rhx-autocomplete.flag

@@ -0,0 +1,2 @@
+branch: vlad/rhx-autocomplete
+bound_by: init.behavior skill

.behavior/v2026_02_22.rhx-autocomplete/.ref.[feedback].v1.[given].by_human.md

@@ -0,0 +1,27 @@
+emit your response to the feedback into
+- .behavior/v2026_02_22.rhx-autocomplete/$BEHAVIOR_REF_NAME.[feedback].v$FEEDBACK_VERSION.[taken].by_robot.md
+
+1. emit your response checklist
+2. exec your response plan
+3. emit your response checkoffs into the checklist
+
+---
+
+first, bootup your mechanics briefs again
+
+npx rhachet roles boot --repo ehmpathy --role mechanic
+
+---
+---
+---
+
+
+# blocker.1
+
+---
+
+# nitpick.2
+
+---
+
+# blocker.3

.behavior/v2026_02_22.rhx-autocomplete/.route/.bind.vlad.rhx-autocomplete.flag

@@ -0,0 +1,2 @@
+branch: vlad/rhx-autocomplete
+bound_by: route.bind skill

.behavior/v2026_02_22.rhx-autocomplete/.route/.drive.blockers.latest.json

@@ -0,0 +1,4 @@
+{
+  "count": 2,
+  "stone": "3.3.blueprint.v1"
+}

.behavior/v2026_02_22.rhx-autocomplete/.route/1.vision.guard.judge.i1p1.26153158050804f95daaa2a463c58e5ac64c9dbe418e9a10bcb7688944875a05.cbee55fb28ea80ec343782b2b20f08377cf1d3a87dbbfcd673e9785fcde5b946.j1.md

@@ -0,0 +1,12 @@
+
+🪨 run solid skill repo=bhrain/role=driver/skill=route.stone.judge
+
+passed: false
+reason: wait for human approval
+
+
+---stderr---
+
+🪨 run solid skill repo=bhrain/role=driver/skill=route.stone.judge
+   └─ ✋ blocked by constraints
+

.behavior/v2026_02_22.rhx-autocomplete/.route/1.vision.guard.judge.i1p1.26153158050804f95daaa2a463c58e5ac64c9dbe418e9a10bcb7688944875a05.e5b912da11ed0245fce121b096c94440006698351e1313c09ad35c2cff6a6b70.j1.md

@@ -0,0 +1,5 @@
+
+🪨 run solid skill repo=bhrain/role=driver/skill=route.stone.judge
+
+passed: true
+reason: human approval found