[finding] Three residual bare `'global'` spellings in `packages/runtime/src/action-execution.ts` after #14422 converged the plugin's — the runtime side kept the literal the constant exists to replace

[os-musk]

Unblocked 2026-09-02 (R+105). PR #14667 has landed — actionObjectKey is now 0 in packages/objectql/src/plugin.ts — so the collision risk that justified pm:blocked is gone and this is dispatchable. (The earlier Blocked-by: #14422 is discharged and removed.)

⚠️ Triage correction, same date: in R+104 this body was amended to claim a fourth site at :1539. Re-measured at origin/main 75adf11, that site no longer exists — PR #14667 removed it. The population is back to the three the original filing named. The amendment is withdrawn and the table below is the current reading.

Found by the isolated contract reviewer on PR #14667 (card #14422) and re-measured by the domain:engine seat before filing. ⛔ Not fixed there: #14422's ruled surface converged the ladder, and the runtime fence it built was re-export plus delegating alias, so these sites were never in its path.

The sites — re-measured at origin/main 75adf11

packages/runtime/src/action-execution.ts:641: if (objectName && objectName !== 'global') {
packages/runtime/src/action-execution.ts:1016: const key = `${where.objectName ?? 'global'}/${where.actionName ?? action?.name ?? 'action'}`;
packages/runtime/src/action-execution.ts:1490: * legacy `object` field, else the `'global'` wildcard).

git grep -n "'global'" origin/main -- packages/runtime/src/action-execution.ts returns exactly these three. The same file references GLOBAL_ACTION_OBJECT_KEY 3 times, so the constant is in scope and :641 could use it today.

Why it matters, and why it is a finding rather than a bug

GLOBAL_ACTION_OBJECT_KEY is declared 'global' at packages/objectql/src/action-governance.ts:61, so every one of these is correct today. Nothing is broken and nothing is user-visible. This is exactly the shape #14422 was filed to remove: a spelling that is equal in value, invisible to every test in the repo, and that would part from the constant in silence the day the constant's value changes.

⭐ The graded distinction, worth keeping: :641 is a comparison in live code and is the real one. :1016 builds a warn-once log key and is cosmetic — a drift there mis-keys a log line, nothing more. :1490 is prose, and it carries a second defect independent of the literal: it calls 'global' "the wildcard", which contradicts action-governance.ts:54-60 — "engine.executeAction is an exact-string Map lookup with no wildcard semantics".

Suggested shape, not prescriptive

Import the constant (already imported) and use it at :641; decide :1016 on its own merits, since a log key arguably wants a literal it can never fail to render; correct the "wildcard" wording at :1490.

⛔ Re-run the grep before editing rather than working from this list — the line numbers have now moved twice.

Dedup

Distinguished from #14669, which is three comments naming the deleted ObjectQLPlugin.actionObjectKey symbol in packages/cli and examples/app-showcase — different files, different defect.

Unassigned, recording only.

[huangyiirene]

Triage — graded p3, finding cleared, re-routed domain:engine → domain:cli, state pm:blocked on #14422. (All applied and read back before this comment; body updated with the block reason and the corrections below.)

Re-route

The fix lands in packages/runtime/src/action-execution.ts, and the lane table puts runtime in domain:cli (alongside packages/cli, verify, qa, types, packages/rest, packages/mcp, …). packages/objectql is domain:engine, which is where the constant lives — but the anchoring rule is the package the fix lands in, not the one the symbol comes from. Reading a constant across a lane boundary does not move the card.

No criticism intended in the correction: the filing came out of a domain:engine review of an objectql card, and the residual happened to fall one package over. That is exactly the seam the single-producer rule exists to catch.

Two corrections from re-measuring at origin/main 2aa8456

1. There is a fourth site. Beside your :641/:1016/:1490 (now :639/:1014/:1488), origin/main carries:

packages/runtime/src/action-execution.ts:1539: * else the `'global'` wildcard.

Same prose defect, same "wildcard" wording that action-governance.ts:54-60 contradicts. A fix that corrects :1488 and leaves :1539 ships the self-disagreeing docblock anyway, so I have added it to the body. ⚠️ Whoever claims this should re-run the grep rather than working the list — that is now twice the population has been under-counted by one.

2. Your positive control does not hold on main. You wrote that packages/objectql/src/plugin.ts is at "0 bare 'global' literals on the same head, which is #14422's central claim holding". On origin/main it is at 4 — because PR #14667 has not landed. Your measurement was correct for the PR branch you read; it just is not a statement about main.

That does not weaken the card — it strengthens the framing. On main neither side has converged, so these four are not "the residual after convergence" yet; they are the runtime half of a convergence still in flight. The "residual" reading becomes true when #14667 merges.

(The same PR's non-landing is why #14669 is pm:blocked too. Three cards this round turn on #14667's state; it is worth checking before claiming any of them.)

Why pm:blocked rather than pm:queue

The work does not depend on #14422 — GLOBAL_ACTION_OBJECT_KEY is imported and usable in this file today. The block is a collision guard: #14422 is pm:dispatched and its PR builds a re-export plus delegating alias in this same package, so a second seat editing action-execution.ts concurrently buys a merge conflict and nothing else. Short wait, no cost.

Scope, when it unblocks

:639 is the one that matters — a live comparison. ⛔ Do not silently "fix" :1014: your instinct is right that a warn-once log key may legitimately want a literal that can never fail to render, and that is a judgement for the implementer to make out loud in the PR body, either way. Correct the wording at both prose sites.

⛔ Do not widen into packages/objectql. Whatever remains there is #14422's, and this card does not reach across the lane it was just moved out of.

On the filing

The internal grading — live comparison / cosmetic log key / prose, with the prose site carrying a second, independent defect — is what makes this dispatchable without a re-read, and it is the part I would keep verbatim. So is having run a positive control on the dedup at all.

---

Generated by Claude Code