instar 1.3.403 → 1.3.405

package/src/data/builtin-manifest.json

@@ -1,8 +1,8 @@
 {
   "$schema": "./builtin-manifest.schema.json",
   "schemaVersion": 1,
-  "generatedAt": "2026-06-07T08:45:36.629Z",
-  "instarVersion": "1.3.403",
+  "generatedAt": "2026-06-07T17:10:48.879Z",
+  "instarVersion": "1.3.405",
   "entryCount": 199,
   "entries": {
     "hook:session-start": {
@@ -11,7 +11,7 @@
       "domain": "identity",
       "sourcePath": "src/core/PostUpdateMigrator.ts",
       "installedPath": ".instar/hooks/instar/session-start.sh",
-      "contentHash": "3e0828b523c2cece3ea4c3c73e77867c2d6884b7493c93e523318e63f5c09e30",
+      "contentHash": "91751f5d379223d5219959314caaeff35e6534ade7acf5193b5768bc1fc79198",
       "since": "2025-01-01"
     },
     "hook:dangerous-command-guard": {
@@ -20,7 +20,7 @@
       "domain": "safety",
       "sourcePath": "src/core/PostUpdateMigrator.ts",
       "installedPath": ".instar/hooks/instar/dangerous-command-guard.sh",
-      "contentHash": "3e0828b523c2cece3ea4c3c73e77867c2d6884b7493c93e523318e63f5c09e30",
+      "contentHash": "91751f5d379223d5219959314caaeff35e6534ade7acf5193b5768bc1fc79198",
       "since": "2025-01-01"
     },
     "hook:grounding-before-messaging": {
@@ -29,7 +29,7 @@
       "domain": "safety",
       "sourcePath": "src/core/PostUpdateMigrator.ts",
       "installedPath": ".instar/hooks/instar/grounding-before-messaging.sh",
-      "contentHash": "3e0828b523c2cece3ea4c3c73e77867c2d6884b7493c93e523318e63f5c09e30",
+      "contentHash": "91751f5d379223d5219959314caaeff35e6534ade7acf5193b5768bc1fc79198",
       "since": "2025-01-01"
     },
     "hook:compaction-recovery": {
@@ -38,7 +38,7 @@
       "domain": "identity",
       "sourcePath": "src/core/PostUpdateMigrator.ts",
       "installedPath": ".instar/hooks/instar/compaction-recovery.sh",
-      "contentHash": "3e0828b523c2cece3ea4c3c73e77867c2d6884b7493c93e523318e63f5c09e30",
+      "contentHash": "91751f5d379223d5219959314caaeff35e6534ade7acf5193b5768bc1fc79198",
       "since": "2025-01-01"
     },
     "hook:external-operation-gate": {
@@ -47,7 +47,7 @@
       "domain": "safety",
       "sourcePath": "src/core/PostUpdateMigrator.ts",
       "installedPath": ".instar/hooks/instar/external-operation-gate.js",
-      "contentHash": "3e0828b523c2cece3ea4c3c73e77867c2d6884b7493c93e523318e63f5c09e30",
+      "contentHash": "91751f5d379223d5219959314caaeff35e6534ade7acf5193b5768bc1fc79198",
       "since": "2025-01-01"
     },
     "hook:deferral-detector": {
@@ -56,7 +56,7 @@
       "domain": "safety",
       "sourcePath": "src/core/PostUpdateMigrator.ts",
       "installedPath": ".instar/hooks/instar/deferral-detector.js",
-      "contentHash": "3e0828b523c2cece3ea4c3c73e77867c2d6884b7493c93e523318e63f5c09e30",
+      "contentHash": "91751f5d379223d5219959314caaeff35e6534ade7acf5193b5768bc1fc79198",
       "since": "2025-01-01"
     },
     "hook:self-stop-guard": {
@@ -65,7 +65,7 @@
       "domain": "coherence",
       "sourcePath": "src/core/PostUpdateMigrator.ts",
       "installedPath": ".instar/hooks/instar/self-stop-guard.js",
-      "contentHash": "3e0828b523c2cece3ea4c3c73e77867c2d6884b7493c93e523318e63f5c09e30",
+      "contentHash": "91751f5d379223d5219959314caaeff35e6534ade7acf5193b5768bc1fc79198",
       "since": "2025-01-01"
     },
     "hook:post-action-reflection": {
@@ -74,7 +74,7 @@
       "domain": "evolution",
       "sourcePath": "src/core/PostUpdateMigrator.ts",
       "installedPath": ".instar/hooks/instar/post-action-reflection.js",
-      "contentHash": "3e0828b523c2cece3ea4c3c73e77867c2d6884b7493c93e523318e63f5c09e30",
+      "contentHash": "91751f5d379223d5219959314caaeff35e6534ade7acf5193b5768bc1fc79198",
       "since": "2025-01-01"
     },
     "hook:external-communication-guard": {
@@ -83,7 +83,7 @@
       "domain": "safety",
       "sourcePath": "src/core/PostUpdateMigrator.ts",
       "installedPath": ".instar/hooks/instar/external-communication-guard.js",
-      "contentHash": "3e0828b523c2cece3ea4c3c73e77867c2d6884b7493c93e523318e63f5c09e30",
+      "contentHash": "91751f5d379223d5219959314caaeff35e6534ade7acf5193b5768bc1fc79198",
       "since": "2025-01-01"
     },
     "hook:scope-coherence-collector": {
@@ -92,7 +92,7 @@
       "domain": "coherence",
       "sourcePath": "src/core/PostUpdateMigrator.ts",
       "installedPath": ".instar/hooks/instar/scope-coherence-collector.js",
-      "contentHash": "3e0828b523c2cece3ea4c3c73e77867c2d6884b7493c93e523318e63f5c09e30",
+      "contentHash": "91751f5d379223d5219959314caaeff35e6534ade7acf5193b5768bc1fc79198",
       "since": "2025-01-01"
     },
     "hook:scope-coherence-checkpoint": {
@@ -101,7 +101,7 @@
       "domain": "coherence",
       "sourcePath": "src/core/PostUpdateMigrator.ts",
       "installedPath": ".instar/hooks/instar/scope-coherence-checkpoint.js",
-      "contentHash": "3e0828b523c2cece3ea4c3c73e77867c2d6884b7493c93e523318e63f5c09e30",
+      "contentHash": "91751f5d379223d5219959314caaeff35e6534ade7acf5193b5768bc1fc79198",
       "since": "2025-01-01"
     },
     "hook:free-text-guard": {
@@ -110,7 +110,7 @@
       "domain": "safety",
       "sourcePath": "src/core/PostUpdateMigrator.ts",
       "installedPath": ".instar/hooks/instar/free-text-guard.sh",
-      "contentHash": "3e0828b523c2cece3ea4c3c73e77867c2d6884b7493c93e523318e63f5c09e30",
+      "contentHash": "91751f5d379223d5219959314caaeff35e6534ade7acf5193b5768bc1fc79198",
       "since": "2025-01-01"
     },
     "hook:claim-intercept": {
@@ -119,7 +119,7 @@
       "domain": "coherence",
       "sourcePath": "src/core/PostUpdateMigrator.ts",
       "installedPath": ".instar/hooks/instar/claim-intercept.js",
-      "contentHash": "3e0828b523c2cece3ea4c3c73e77867c2d6884b7493c93e523318e63f5c09e30",
+      "contentHash": "91751f5d379223d5219959314caaeff35e6534ade7acf5193b5768bc1fc79198",
       "since": "2025-01-01"
     },
     "hook:claim-intercept-response": {
@@ -128,7 +128,7 @@
       "domain": "coherence",
       "sourcePath": "src/core/PostUpdateMigrator.ts",
       "installedPath": ".instar/hooks/instar/claim-intercept-response.js",
-      "contentHash": "3e0828b523c2cece3ea4c3c73e77867c2d6884b7493c93e523318e63f5c09e30",
+      "contentHash": "91751f5d379223d5219959314caaeff35e6534ade7acf5193b5768bc1fc79198",
       "since": "2025-01-01"
     },
     "hook:stop-gate-router": {
@@ -137,7 +137,7 @@
       "domain": "safety",
       "sourcePath": "src/core/PostUpdateMigrator.ts",
       "installedPath": ".instar/hooks/instar/stop-gate-router.js",
-      "contentHash": "3e0828b523c2cece3ea4c3c73e77867c2d6884b7493c93e523318e63f5c09e30",
+      "contentHash": "91751f5d379223d5219959314caaeff35e6534ade7acf5193b5768bc1fc79198",
       "since": "2025-01-01"
     },
     "hook:auto-approve-permissions": {
@@ -146,7 +146,7 @@
       "domain": "safety",
       "sourcePath": "src/core/PostUpdateMigrator.ts",
       "installedPath": ".instar/hooks/instar/auto-approve-permissions.js",
-      "contentHash": "3e0828b523c2cece3ea4c3c73e77867c2d6884b7493c93e523318e63f5c09e30",
+      "contentHash": "91751f5d379223d5219959314caaeff35e6534ade7acf5193b5768bc1fc79198",
       "since": "2025-01-01"
     },
     "job:health-check": {
@@ -1538,7 +1538,7 @@
       "type": "subsystem",
       "domain": "updates",
       "sourcePath": "src/core/PostUpdateMigrator.ts",
-      "contentHash": "3e0828b523c2cece3ea4c3c73e77867c2d6884b7493c93e523318e63f5c09e30",
+      "contentHash": "91751f5d379223d5219959314caaeff35e6534ade7acf5193b5768bc1fc79198",
       "since": "2025-01-01"
     },
     "subsystem:scheduler": {

package/src/scaffold/templates.ts

@@ -695,6 +695,13 @@ This routes feedback to the Instar maintainers automatically. Valid types: \`bug
 - Returns \`{ available, usage: { primary, secondary, model, planType, rateLimitReachedType } }\` where each window has \`usedPercent\`, \`remainingPercent\`, \`windowMinutes\`, \`resetsAt\`/\`resetsAtIso\`, \`resetsInSeconds\`. \`available:false\` means no codex session data on disk yet (e.g. a pure-Claude agent).
 - **When to use**: when asked "how much codex usage is left?" / "am I near the limit?", before scheduling heavy codex work, or to drive a model-swap when a window is exhausted (\`rateLimitReachedType\` is non-null, or \`secondary.remainingPercent\` is low).
 
+**Subscription Pool (multi-account quota + auto-swap + enrollment)** — Hold ALL of your subscriptions for a provider (e.g. several Claude logins) and use them as one pool: I read each account's live quota, drain each before its reset, and when a session hits an account's limit I resume it on another account instead of letting it die. The registry stores each account's login LOCATION (its config home), NEVER a token.
+- See the pool + each account's live quota: \`curl -H "Authorization: Bearer $AUTH" http://localhost:${port}/subscription-pool\` · one account's quota + burn: \`GET /subscription-pool/:id/quota\` · poll all now: \`POST /subscription-pool/poll\`.
+- **Continuity guarantee** — a long session that hits its account's quota resumes on another eligible account (conversation preserved via \`--resume\`), never dies. Manual lever: \`POST /subscription-pool/swap\` \`{"sessionName":"...","exhaustedAccountId":"..."}\`. Auto-swap on rate-limit ships OFF (opt-in via \`subscriptionPool.autoSwapOnRateLimit\` — it moves a live session, real authority).
+- **Enroll a new account from your phone** — \`POST /subscription-pool/enroll\` \`{"id","label","provider","framework","configHome"}\` starts a login and returns a public code/URL (never a token); \`GET /subscription-pool/pending-logins\` is the surface; expired codes are auto-reissued. Mark done with \`POST /subscription-pool/enroll/:id/complete\`.
+- **Dashboard**: the **Subscriptions tab** shows live quota bars (5h + weekly + reset countdown), status, and the Pending Logins panel — share the dashboard URL + PIN.
+- **When to use** (PROACTIVE): "how much quota is left across my accounts?" / "am I about to hit a limit?" → \`GET /subscription-pool\`; the user wants to add another subscription → drive the enrollment wizard (never ask them to paste a token); a long job is at risk of a quota wall → the continuity guarantee + \`/swap\` keep it alive. Single-account pools are a no-op.
+
 **Per-Feature LLM Metrics** — See what each of your LLM-driven gates/sentinels actually costs and how often it fires, so tuning them is evidence-based (which to thin, which to strengthen). Read-only observability (like token usage) — it never gates anything.
 - Check: \`curl -H "Authorization: Bearer $AUTH" "http://localhost:${port}/metrics/features?sinceHours=24"\`
 - Returns \`{ totals, features: [{ feature, calls, tokensIn, tokensOut, fired, noop, fireRate, p50LatencyMs, p95LatencyMs, ... }] }\` — one row per system (e.g. MessagingToneGate, CoherenceReviewer). Filter with \`?feature=<name>\`.

package/upgrades/1.3.404.md

@@ -0,0 +1,37 @@
+# Upgrade Guide — vNEXT
+
+<!-- assembled-by: assemble-next-md -->
+<!-- bump: patch -->
+
+## What Changed
+
+The Subscription & Auth Standard's `/subscription-pool` routes graduated from the
+internal/dark list to a surfaced capability, now that all its phases are merged —
+the multi-account registry (P1.1), quota poller (P1.2), quota-aware auto-swap
+scheduler with the session-continuity guarantee (P1.3), mobile-first enrollment
+wizard (P2.1), and dashboard Subscriptions tab (P2.2).
+
+Three no-behavior edits: `/subscription-pool` moves from `INTERNAL_PREFIXES` to a
+`CAPABILITY_INDEX` entry (so it appears in `/capabilities` with its account count,
+quota poller / scheduler / enrollment-wizard wiring status, and full endpoint
+list); the CLAUDE.md template gains a "Subscription Pool" awareness blurb for new
+agents; and an idempotent `migrateClaudeMd` migration appends the same blurb to
+existing agents on update. No route, class, or runtime behavior is added — this
+surfaces an already-shipped, already-tested capability honestly (the original
+internal note set exactly this maturity bar).
+
+## What to Tell Your User
+
+I can now manage several of your subscriptions for a provider as one pool — show
+you how much of each is left and when it resets, keep a long session alive by
+moving it to another account when one hits its limit, and help you enroll a new
+account from your phone. Ask me "how much quota is left across my accounts?" or
+tell me you want to add another subscription.
+
+## Summary of New Capabilities
+
+- **`/subscription-pool` is now a surfaced capability** — visible in
+  `/capabilities` and documented in the agent's CLAUDE.md (new + existing agents).
+- **No behavior change** — the routes, scheduler, continuity guarantee, and
+  enrollment wizard already shipped (P1.1–P2.2); this makes the finished capability
+  discoverable and known.

package/upgrades/1.3.405.md

@@ -0,0 +1,44 @@
+# Upgrade Guide — vNEXT
+
+<!-- assembled-by: assemble-next-md -->
+<!-- bump: patch -->
+
+## What Changed
+
+The idle-session reaper checked "is this session working?" with a pattern that matched
+tool-call names (Read(/Write(/Bash() and the bare word "claude" anywhere in the captured
+terminal — but those persist in an IDLE session's scrollback, so every idle Claude
+session read as "working" and the reaper never reaped one (0 reaps in 9.5h live; the
+residual blocker after #961). Fix: `isPositivelyIdle` now uses a new live-only signal
+(`liveActivity`: spinner glyphs + "Working (Ns" + "generating"), which vanish when a turn
+ends), not the scrollback-persistent `toolCallOrSpinner`. Also removed the bare word
+"claude" from the Claude activity pattern (the documented codex "do not match bare
+'codex'" lesson).
+
+## What to Tell Your User
+
+Nothing required — internal reaper classifier; reaper stays opt-in. For operators who
+enable it: the reaper can now correctly recognize a genuinely-idle session and reclaim it
+(it previously could not). The change also makes the silence/presence sentinels stop
+mis-reading idle Claude sessions as "working".
+
+## Summary of New Capabilities
+
+- `FrameworkActivitySignal.liveActivity` — live-generation-only markers per framework, so
+  callers that need "working RIGHT NOW" don't trip over scrollback history.
+- `SessionReaper.isPositivelyIdle` uses `liveActivity`; bare "claude" removed from the
+  Claude `toolCallOrSpinner`.
+
+## Evidence
+
+`tests/unit/session-reaper.test.ts`: idle Claude pane (scrollback tool-names + "claude",
+no live marker) → positively idle; working pane (spinner / spinner+esc) → not idle; bare
+"claude" alone no longer forces not-idle. 57/57 green. `tsc --noEmit` clean. causalAutopsy:
+latent (exposed by the #955/#958/#961 chain reaching the positive-idle check).
+
+## Scope (honest)
+
+Final functional fix in the chain (#952 → #955 → #958 → #961 → this): the reaper can now
+tell idle from working and actually reclaim. Validated by dry-run before any live flip.
+Does NOT address the transcript pile-up (151k files / 5.9 GB) — separate retention
+follow-up.

package/upgrades/side-effects/reaper-live-activity-idle.md

@@ -0,0 +1,46 @@
+# Side-Effects Review - Reaper live-activity idle detection
+
+**Version / slug:** `reaper-live-activity-idle`
+**Date:** `2026-06-07`
+**Author:** `echo`
+**Second-pass reviewer:** `not required (Tier 1)`
+
+## Summary of the change
+
+`SessionReaper.isPositivelyIdle` used `sig.toolCallOrSpinner` to decide "is this session active?", but that regex matches tool-call names (Read(/Write(/Bash(/…) and the bare word "claude" — both of which PERSIST in an idle session's 200-line scrollback. So every idle Claude session read as active → never positively idle → reaper kept everything (0 reaps in 9.5h live; the residual blocker after #961 made transcripts resolve). Adds a `liveActivity` regex to `FrameworkActivitySignal` (live-generation-only: spinner glyphs + "Working (Ns" + "generating") and uses it in `isPositivelyIdle`; removes the bare `claude|` from the Claude `toolCallOrSpinner` (the documented codex "do not match bare 'codex'" lesson).
+
+## Decision-point inventory
+
+- `FrameworkActivitySignal.liveActivity` (new field) on all 4 signals (claude/codex/gemini/pi) — live-only markers.
+- `SessionReaper.isPositivelyIdle`: active-check now `sig.liveActivity || sig.escapeToInterrupt || sig.runningIndicator` (was `sig.toolCallOrSpinner || …`).
+- `CLAUDE_CODE_SIGNAL.toolCallOrSpinner`: bare `claude|` removed (tool-names + spinner retained).
+
+## 1. Behavior change / gating
+
+This makes `isPositivelyIdle` correctly recognize idle sessions; it does not change any other gate. The only behavioral effect: the reaper can now reach reap-eligibility for genuinely-idle sessions (it previously could not). Removing bare "claude" from toolCallOrSpinner also makes OTHER consumers (silence/presence sentinels) stop mis-reading idle Claude sessions as working — strictly more correct, matching the existing codex behavior. No API/route/state change.
+
+## 2. Over/under-signal
+
+Direction of risk: a session genuinely working but whose captured frame momentarily shows no live marker (between tool calls) could read idle. Mitigations: (a) transcript-growth fires first and KEEPS anything whose JSONL is growing — a working session is growing its transcript; (b) confirmObservations (3 ticks / render-stasis) requires sustained idle; (c) 8h-silence + the full guard chain still apply. So a momentary live-marker miss cannot produce a wrong reap. UNDER-signal (the prior "never recognizes idle") is the bug being fixed. `liveActivity` deliberately keeps every genuine live marker per framework (spinner always; +Working-status for codex; +generating/thinking for gemini/pi).
+
+## 3. Blast radius
+
+Pure regex/classifier logic in the shared signal module + one call-site in the reaper. No I/O, no new deps, no persistent state, no migration. Other consumers of `toolCallOrSpinner` are unaffected except they no longer match the bare word "claude" (a correctness improvement, not a regression — no test depended on it; 57 unit tests incl. the framework-signal suite pass).
+
+## 4. Failure modes
+
+`liveActivity` is a required field with a concrete regex on every signal (tsc-enforced — a missing one is a compile error). A malformed frame is just text the regexes test against; no throw path. The reaper's downstream stateful checks (transcript-growth, positive-idle prompt match, confirmObservations) all still run after this gate.
+
+## 5. Migration parity
+
+No agent-installed file changes; internal classifier logic shipped in code, effective on next server start. No config surface. The reaper remains opt-in + dry-run-first, so nothing changes for an operator until they enable it.
+
+## 6. Scope honesty (what this is NOT)
+
+- Final functional fix in the chain (#952 → #955 → #958 → #961 → this). With it the reaper can distinguish idle from working and actually reclaim; validated by dry-run before any live flip.
+- Does NOT touch the gemini/codex/pi live-marker sets beyond adding the live-only subset (codex/gemini/pi toolCallOrSpinner were already largely live-only or carefully tuned; only Claude carried the bare-word + tool-name-in-idle bug at the reaper call-site).
+- Does NOT address the transcript pile-up (151k files / 5.9 GB) — separate retention follow-up.
+
+## 7. Causal autopsy
+
+Origin: **latent**. The reaper's idle-check has used the scrollback-matching `toolCallOrSpinner` (incl. bare "claude") since `isPositivelyIdle` was written — always wrong for the "is it active NOW" question, but never hit because the earlier KEEP-guards short-circuited before the positive-idle check ran. The 2026-06-07 #955/#958/#961 chain peeled those layers (and #961 made transcripts resolve), so the reaper finally REACHED the positive-idle check and the latent bug surfaced as 0 reaps / universal `no-positive-idle`. No prior PR regressed it; it is the deepest layer of the same conservatism stack. The bare-"claude" half is the exact analogue of the already-fixed codex bare-word bug.

package/upgrades/side-effects/subscription-auth-graduate.md

@@ -0,0 +1,72 @@
+# Side-Effects Review — Graduate the Subscription Pool capability (P-graduate)
+
+## Scope of change
+
+- `src/server/CapabilityIndex.ts` — remove `subscription-pool` from `INTERNAL_PREFIXES`; add a `subscriptionPool` entry to `CAPABILITY_INDEX` (key, prefixes, description, build → configured/accounts/quotaPoller/scheduler/enrollmentWizard/endpoints).
+- `src/scaffold/templates.ts` (`generateClaudeMd`) — add the Subscription Pool awareness blurb so NEW agents know the capability.
+- `src/core/PostUpdateMigrator.ts` (`migrateClaudeMd`) — add a content-sniffed, idempotent migration so EXISTING agents get the same blurb on update (Migration Parity).
+- `upgrades/next/subscription-auth-graduate.md` — release-note fragment.
+
+## What this does (and does NOT) change
+
+This is a **classification + awareness** change. It surfaces an already-shipped,
+already-tested capability (P1.1 registry, P1.2 poller, P1.3 scheduler + continuity
+guarantee, P2.1 enrollment, P2.2 dashboard) to `/capabilities` and to the agent's
+CLAUDE.md. It adds **no route, no class, no runtime behavior** — the routes, the
+scheduler, the wizard already exist and are wired. The `build()` function is pure
+(reads `ctx.subscriptionPool` etc. already on the context) and runs only when
+`/capabilities` is probed.
+
+## Why now (maturity honesty)
+
+The original `INTERNAL_PREFIXES` entry explicitly said `subscription-pool`
+"graduates to a surfaced CAPABILITY_INDEX entry (+ CLAUDE.md awareness blurb) once
+the quota-aware scheduler (P1.3) and mobile enrollment wizard (P2.1) make it
+user-usable. Surfacing the bare registry now would overclaim an unfinished
+capability." P1.3 + P2.1 (+ P2.2 dashboard) are now merged — the maturity bar the
+note set is met, so surfacing it is honest, not overclaiming.
+
+## Authority / autonomy analysis
+
+- **No new authority.** The auto-swap of live sessions remains OFF by default
+  (`subscriptionPool.autoSwapOnRateLimit`); the blurb documents it as opt-in. The
+  enrollment blurb tells the agent to drive the wizard and NEVER ask the user to
+  paste a token — reinforcing the existing credential-safety posture.
+- **The migration edits agent-installed CLAUDE.md** — but only appends a doc
+  section, content-sniffed on a distinctive marker (idempotent: re-running is a
+  no-op), and never touches a custom/user-authored section. This is the standard
+  `migrateClaudeMd` append pattern used by ~30 prior sections.
+
+## Failure modes considered
+
+- Capability probed on an agent with no pool wired → `build()` returns
+  `{ configured:false, accounts:0, ... }` (no throw; the routes already answer
+  `200 { enabled:false }`).
+- Migration run twice → content-sniff marker present → skipped (idempotent).
+- Migration on an agent whose CLAUDE.md predates the section → appended once.
+- The capabilities-discoverability lint: `/subscription-pool` is now claimed by
+  exactly one CAPABILITY_INDEX entry and absent from INTERNAL_PREFIXES (verified:
+  CapabilityIndex.test.ts green).
+
+## Blast radius
+
+Minimal. Classification metadata + a documentation blurb + an idempotent doc
+migration. No behavior path changes; `/capabilities` gains one block; new + existing
+agents gain one CLAUDE.md section.
+
+## Framework generality
+
+Framework-agnostic in the surface, Claude-first in the wired mechanism (matching the
+standard's scope, Justin's decision 3): the CAPABILITY_INDEX entry + blurb describe
+the pool generically (provider/framework-parameterized), and the enrollment + quota
+machinery already carry `provider`/`framework` fields. The continuity-guarantee swap
+is implemented for `claude-code` today (per-account `CLAUDE_CONFIG_DIR`); other
+frameworks slot into the same registry + selector when their per-account config is
+wired. Surfacing the capability does not bake in any Claude-only assumption beyond
+what the underlying (already-merged) phases already chose.
+
+## Migration / parity
+
+This change IS the parity work: `generateClaudeMd` (new agents) + `migrateClaudeMd`
+(existing agents) both get the blurb, satisfying the Agent Awareness + Migration
+Parity standards. No config/hook/skill change. Ships via dist.