@etiquekit/etq 1.0.10 → 1.0.11

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@etiquekit/etq",
-  "version": "1.0.10",
+  "version": "1.0.11",
   "license": "Apache-2.0",
   "description": "Governance and evidence CLI for AI-assisted engineering seats.",
   "private": false,

package/templates/DISPATCH_PREMISES.md

@@ -0,0 +1,39 @@
+# Dispatch Premises (Load-Bearing Premise Gate)
+
+Reusable discipline block for dispatch cards. A load-bearing premise is any
+claim whose falsity changes the dispatch's verdict, scope, or architecture.
+Refuted premises waste a full runner round trip; the gate moves the check to
+the dispatcher, where it is cheapest.
+
+## Gate Rules
+
+1. If a load-bearing premise is statically decidable at the routed base, the
+   dispatcher must VERIFY it before posting. Proof means a file:line whose
+   CONTENTS were read at the pinned base, on the CONSUMING surface's path.
+   Producer-side or sibling-binary evidence never verifies a consumer claim.
+   Name or path existence is never content.
+2. Premises decidable only at runtime or by an external system ship as a
+   labeled HYPOTHESIS with a paired stop-condition the runner can trigger.
+3. Budget: about 15 minutes and at most 5 premises. Beyond that, downgrade
+   the dispatch to a probe task whose deliverable IS the premise table.
+4. Lint tell: a stop-condition that restates a static premise verbatim is a
+   fence instead of a check — verify it at dispatch time instead.
+5. Runner stop-conditions remain the backstop, not the primary check. A
+   runner that finds a premise false must STOP and counter, never fabricate
+   a passing result.
+
+## Template Block
+
+Paste into the dispatch card and fill in:
+
+```markdown
+## Premises
+
+| # | premise | class | proof / stop-condition |
+| --- | --- | --- | --- |
+| P1 | <claim the dispatch depends on> | VERIFIED | <file:line, contents read at <base-sha>, on the consuming surface> |
+| P2 | <claim decidable only at runtime> | HYPOTHESIS | STOP if <observable falsifier>; counter instead of fabricating a pass |
+```
+
+Every load-bearing premise appears in the table. A dispatch with an
+unverified static premise is not ready to post.

package/templates/etiquette-vanilla-v0/source/control-seat/docs/work/access/ACCESS-ASSURANCE.md

@@ -24,7 +24,7 @@ Provider-backed secure deployments should keep raw identity material outside the
 repo:
 
 ```text
-Persona / Scandit / SSO -> Supabase derived claims -> local access check
+approved identity provider / assurance provider -> derived claims -> local access check
 ```
 
 Repo-safe fields are provider refs, status, policy eligibility class, evidence

package/templates/etiquette-vanilla-v0/source/control-seat/docs/work/access/workspace-secure-profile.v0.json

@@ -23,7 +23,7 @@
   },
   "claim_source": {
     "kind": "none-required-for-open-mode",
-    "live_replacement": "Supabase Auth plus derived Persona and Scandit claim export",
+    "live_replacement": "approved identity provider plus derived assurance claim export",
     "raw_pii_allowed_in_repo": false
   },
   "reactor": {

package/templates/seat-packs-v0/source/claude-code-seat/README.md

@@ -17,3 +17,29 @@ Before pickup in the target project, read:
 Re-orient with `etiquette whereami` before pickup after stale context, resume,
 runner/model/profile swaps, or workflow migration. Hooks and readiness are
 evidence only; the task envelope is the work authority.
+
+## Claude Code QoL
+
+**Monitoring posture (push vs pull):**
+
+- Default OFF. A solo seat has no counterpart returns to catch — poll on
+  demand at natural checkpoints (`etiquette notify poll --seat <seat>` after
+  dispatching, before gates) instead of arming a watcher.
+- Arm the Monitor tool (push) only when counterpart seats are actively
+  working or during a coordination window. Wrap a cursor-based poll loop —
+  never `tail -F` an event file from byte 0 (git operations rewrite it and
+  replay history).
+- Push and the interactive agents view are mutually exclusive: any
+  persistent background task blocks the agents shortcut. Toggle deliberately
+  and say which posture you are in.
+
+**Context-compaction hazards (Claude Code specific):**
+
+- Compaction can silently swap your model. Render the active model in your
+  statusline; verify it after every compaction or resume.
+- Re-orient with `etiquette whereami` after compaction — chat memory is not
+  a task binding. Durable facts belong in files, not context.
+
+**Continuity:** prefer forked subagents for work that must survive your
+session; a plain subagent dies with you. Keep persistent background tasks to
+a minimum — each one is a posture commitment.

package/templates/seat-packs-v0/source/codex-seat/README.md

@@ -20,3 +20,17 @@ evidence only; the task envelope is the work authority.
 
 For bounded, cursor-based wake checks use the portable core command
 `etiquette notify poll --seat <id> --after-seq <n>` instead of raw `tail -F` monitors.
+
+## Codex QoL
+
+**Runner claims:** one seat, one active runner per task. Before occupying a
+seat another runner may hold, claim explicitly and record your runner id; if
+you find another runner's work in the worktree, preserve it and stand down or
+counter — never overwrite.
+
+**Honest substitution:** when your sandbox/safety layer refuses a declared
+command (e.g. a disable-sandbox flag), run the stricter variant and disclose
+the substitution in your return — never silently claim the declared gate ran.
+
+**Notifications:** use bounded cursor-based polls (`etiquette notify poll`);
+pair long waits with an explicit re-check cadence in your session notes.

package/templates/seat-packs-v0/source/gemini-seat/README.md

@@ -17,3 +17,23 @@ Before pickup in the target project, read:
 Re-orient with `etiquette whereami` before pickup after stale context, resume,
 runner/model/profile swaps, or workflow migration. Hooks and readiness are
 evidence only; the task envelope is the work authority.
+
+## Gemini QoL
+
+**Role fit:** this seat shines as a divergent-advisory and review seat —
+fast structured review-returns (`confirm_with_amendments` with concrete
+amendments), coverage sweeps, and alternative-angle pressure tests. Treat
+its returns as first-class evidence.
+
+**Calibration duties:**
+
+- Every claim in a return carries file:line evidence read on the consuming
+  surface — speed invites overclaim; evidence is the antidote.
+- Do not act as the sole verifier for promotion-grade claims; pair with a
+  second party or an executable check.
+- Keep ledger-facing writes in a plain technical register — schemas, fields,
+  checks; metaphors stay in chat.
+
+**Notifications:** Claude-style push monitors do not port to this runtime.
+Use bounded cursor-based polls (`etiquette notify poll --seat <seat>`); no
+persistent watchers, no tail-based loops.