feat(sdk): make project context integrity explicit

.agents/pm/extensions/.managed-extensions.json

@@ -1,12 +1,12 @@
 {
   "version": 1,
-  "updated_at": "2026-07-27T06:24:07.945Z",
+  "updated_at": "2026-07-27T08:42:53.617Z",
   "entries": [
     {
       "name": "pm-changelog",
       "directory": "pm-changelog",
       "scope": "project",
-      "manifest_version": "2026.7.26",
+      "manifest_version": "2026.7.27",
       "manifest_entry": "./dist/extension.js",
       "capabilities": [
         "commands",
@@ -15,13 +15,13 @@
         "schema"
       ],
       "installed_at": "2026-05-23T20:26:03.509Z",
-      "updated_at": "2026-07-26T07:08:27.220Z",
+      "updated_at": "2026-07-27T08:18:38.283Z",
       "source": {
         "kind": "npm",
         "input": "npm:pm-changelog",
         "location": "package",
         "package": "pm-changelog",
-        "version": "2026.7.26"
+        "version": "2026.7.27"
       }
     }
   ]

.agents/pm/features/pm-w7mqzt.toon

@@ -6,20 +6,20 @@ status: open
 priority: 2
 tags[5]: "area:extensions","area:sdk",customization,dx,refactor
 created_at: "2026-07-25T21:40:27.254Z"
-updated_at: "2026-07-26T14:09:14.479Z"
+updated_at: "2026-07-27T07:52:09.950Z"
 author: "harness:claude-code"
 estimated_minutes: 420
 acceptance_criteria: "An extension can declare the standard filter-flag set by consuming an SDK export rather than by restating FlagDefinition objects, and adding a filter flag to the CLI baseline reaches consuming packages without per-package edits. The SDK provides the runtime-resolution helper that the ten first-party copies of ensureRuntimeBundle currently implement, so no package hand-joins dist paths; the first-party copies are replaced by it. Alias and cursor semantics attached to a baseline flag are inherited rather than re-declared, and pm-search-advanced's divergent aliases are either adopted into the baseline or dropped. At least one first-party package is migrated end to end as proof and its hand-written tables are deleted. A regression test asserts that a package built only against public SDK subpaths can register a command carrying the full baseline filter set."
 parent: pm-ugqx
 risk: medium
 confidence: high
-dependencies[5]{id,kind,created_at,author,source_kind}:
-  pm-9rgaal,implements,"2026-07-25T21:40:27.254Z",null,null
-  pm-9smp7j,related,"2026-07-25T21:40:27.254Z",null,null
-  pm-c5f0gh,related,"2026-07-25T21:40:27.254Z",null,null
-  pm-usfg,related,"2026-07-25T21:40:27.254Z",null,null
-  pm-x6jf,related,"2026-07-25T21:40:27.254Z",null,null
-comments[20]{created_at,author,text}:
+dependencies[5]{id,kind,created_at}:
+  pm-9rgaal,implements,"2026-07-25T21:40:27.254Z"
+  pm-9smp7j,related,"2026-07-25T21:40:27.254Z"
+  pm-c5f0gh,related,"2026-07-25T21:40:27.254Z"
+  pm-usfg,related,"2026-07-25T21:40:27.254Z"
+  pm-x6jf,related,"2026-07-25T21:40:27.254Z"
+comments[22]{created_at,author,text}:
   "2026-07-25T22:20:08.935Z","harness:codex","Start evidence: all-status inventory reviewed 2,043 items (1,766 closed, 265 open, 12 canceled) with zero pre-existing in-progress items. Selected one cohesive package-authoring lane spanning consumable SDK flag/runtime contracts, scaffold parity, prefix-safe naming, and dependency queue reconciliation. GitHub snapshot has zero open PRs; all 21 open issues contain canonical pm-item links; pnpm audit reports zero vulnerabilities."
   "2026-07-25T22:35:02.712Z","harness:codex","Implementation evidence: added public CLI-contract to extension-flag adapter, exported list/search baselines, migrated pm-search-advanced through the first-party runtime bridge, and passed 158 focused unit/integration tests plus targeted ESLint."
   "2026-07-25T22:38:02.233Z","harness:codex","Contract evidence: full CLI contract snapshot stayed current; SDK surface update classified eight additive exports across sdk, sdk/authoring, and sdk/runtime with no breaking or changed symbols."
@@ -40,6 +40,8 @@ comments[20]{created_at,author,text}:
   "2026-07-26T13:36:38.633Z","harness:codex","Hosted lint evidence: PR #744 head 47fc3771 reports obsolete RuntimeSdkModule in governance-audit public runtime migration."
   "2026-07-26T13:58:40.912Z","harness:codex","Correction evidence: removed the obsolete RuntimeSdkModule compatibility interface from governance-audit and now derives the runtime contract directly from typeof runtime.runList. Focused package runtime coverage and repository lint pass; final sandboxed coverage completed 374 files and 6,420 tests at exact 100/100/100/100. Remaining first-party dynamic loader convergence is still outstanding, so returning this item to open."
   "2026-07-26T14:09:14.478Z","harness:codex","Exact-head hosted evidence: PR #744 commit b28d3cb736a32cbcd54266243cd3f6848372f6df passed the complete required GitHub check matrix with zero failed checks. Thread-aware inventory found zero inline threads. CodeRabbit completed the explicitly retriggered full-review request under its external fair-usage limit and emitted no finding; all new bot surfaces were reacted to and acknowledged. Local final evidence remains 374 files, 6,420 tests, exact 100/100/100/100 coverage, clean lint/static/contracts/changelog/health, and real-time telemetry drain."
+  "2026-07-27T07:35:20.721Z","harness:codex","Start evidence 2026-07-27: selected after live all-status review of 2,129 items (1,807 closed, 309 open, 13 canceled; zero pre-existing in-progress), full canonical metadata/closed-lineage review, and GitHub/security/dependency inventory. This item is actively implemented in one cohesive public SDK/CLI integrity PR; unrelated backlog and container epics remain open and unclaimed."
+  "2026-07-27T07:52:07.934Z","harness:codex","Scope correction 2026-07-27: live implementation showed the remaining runtime-loader convergence requires a dedicated public resolution primitive plus migration of package entrypoint topology and its packed-install matrix. This cohesive PR is already broad across five completed SDK/CLI integrity issues; no runtime-loader source change is being shipped here, so returning this item to open and releasing ownership rather than misrepresenting it as active."
 notes[1]{created_at,author,text}:
   "2026-07-26T11:33:06.917Z","harness:claude-code","Positive control on the scaffold path, run 2026-07-26 on the shipped 2026.7.26 build, recorded to bound this item's scope precisely.\n\nThree commands take a package from nothing to a working command, with no npm install and no build step:\n\n  pm package --init demo-tracker      -> 7 files (package.json, manifest.json, index.ts, index.test.ts, tsconfig.json, .gitignore, README.md)\n  pm install --project ./demo-tracker -> activation_status ok, activated true, registered_commands ['demo tracker ping'], registered_actions ['demo-tracker-ping'], health ok, 0 blocking failures\n  pm demo tracker ping                -> ok: true, message 'Starter extension scaffold is active.'\n\nThe native type-stripping loader means a .ts entry runs directly, so the scaffold's promise is real and the getting-started experience is genuinely three commands. The next_steps block returned by --init lists install, smoke test, typecheck, test, and doctor in the right order.\n\nThe consequence for this item: the gap is not onboarding, it is the second step. A package author who accepts the generated shape is well served; one who needs a real flag surface hits the missing adapter between LIST_FILTER_FLAG_CONTRACTS/CliFlagContract and registerCommand's FlagDefinition that this item owns. Framing the work as 'the scaffold stops one layer short of the contracts the CLI itself uses' is more accurate than 'authoring is hard', and it keeps the fix scoped to the adapter rather than to the generator.\n\nIncidental observation, not a defect claim: a hyphenated target produces a two-token command path (demo-tracker -> 'pm demo tracker'). That is consistent with the noun-verb grammar; the pm-prefixed variant of the same derivation is pm-c5f0gh's subject."
 files[17]{path,scope,note}: