fix(server): Codex V2 runs skills typed with any currency sigil

[juliusmarminge]

The web and mobile composers let users trigger a skill with any currency symbol, not just $ (#12098), so €review shows up as a skill chip. V1 Codex rewrote the sigil to $ before sending the turn. The V2 Codex adapter sent the text unchanged. Codex only parses $name as a skill mention (TOOL_MENTION_SIGIL in codex-rs/skills/src/mentions.rs at rust-v0.156.1), so on V2 €review reached Codex as plain text and the skill was never invoked.

What changed

• packages/shared/src/composerInlineTokens.ts now exports SKILL_MENTION_PATTERN, the prompt-side skill mention regex. The composer's chip regex is built from the same source. The two differ only at the end: the chip regex needs whitespace after the name, so a half-typed $re at the cursor doesn't become a chip. The prompt pattern also matches at end of text, because a sent prompt can end with a mention. This is the same (?=\s|$) variant that V1 Codex, Claude and Cursor already copy locally.
• CodexAdapterV2.ts adds codexSkillMentionText, which does the same $1$$$2 rewrite as V1. toCodexInput applies it to the user's message text before attachment context is appended. That covers both turn/start and turn/steer. Currency amounts (€20, €5k, €100M, €1e6) and glued tokens (5€review) don't match and stay prose.

The Claude and Cursor drivers and the V1 runtime still have their own local copies of the pattern. I left them alone to keep this PR focused, and because V1 is being removed separately.

Verification

• vp test run src/orchestration-v2/Adapters/CodexAdapterV2.test.ts src/orchestration-v2/testkit/CodexReplayFixtures.integration.test.ts (apps/server): 118 passed. The new table test covers €review do it → $review do it, £ship, a mid-sentence ¥review, a mention after a newline, the astral 𑿝, digit-leading €2spec, unchanged $review, and prose left alone: costs €20, €5k, €100M/€1e6, 5€review.
• A scripted-replay test runs startTurn with €review do it and steerTurn with then £ship it, and expects the outbound turn/start and turn/steer frames to carry $review do it and then $ship it. Revert check: with toCodexInput reverted to text: turnInput.message.text, this test fails with CodexAppServerReplayFrameMismatchError on turn/start (received €review do it), and it passes with the fix.
• None of the Codex replay fixtures has a turn/start or turn/steer input with a currency-prefixed token, so no transcript changes. The fixture suite passes unchanged.
• vp test run src/composerInlineTokens.test.ts src/composerTrigger.test.ts (packages/shared): 45 passed. vp test run src/composer-editor-mentions.test.ts (apps/web): 34 passed. These cover the chip regex that is now derived from the shared source.
• tsc --noEmit for apps/server and packages/shared: no errors. vp lint on the touched files: no new warnings.
• Live check with codex exec --ephemeral 0.156.1 in a temp repo with a .agents/skills/t3probe skill. For $t3probe, Codex injected the skill and the model answered with the skill's marker directly, with no tool calls. For €t3probe, the skill was not injected. The model guessed the intent from the skill list and ran sed to read the SKILL.md itself (it also picked up an unrelated user skill). So € only works when the model happens to guess, which confirms $ is the only sigil Codex parses.
• Not run: a full orchestrator run of the V2 Codex adapter against a live provider.

Model: Claude Opus 5.5 (Claude Code)

🤖 Generated with Claude Code

---

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

ℹ️ No successful main baseline artifact is available yet. This run establishes the initial measurement.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	—	4.9 KiB	—	6.8 KiB	✅
Codex	Thread snapshot wire	—	3.7 KiB	—	4.9 KiB	✅
Codex	Live turn WebSocket wire	—	1.1 KiB	—	2.0 KiB	✅
Codex	Live turn WebSocket decoded	—	20.4 KiB	—	29.3 KiB	✅
Codex	Live turn messages	—	1	—	8	✅
Claude	Total thread wire	—	4.9 KiB	—	6.8 KiB	✅
Claude	Thread snapshot wire	—	3.7 KiB	—	4.9 KiB	✅
Claude	Live turn WebSocket wire	—	1.2 KiB	—	2.0 KiB	✅
Claude	Live turn WebSocket decoded	—	20.8 KiB	—	29.3 KiB	✅
Claude	Live turn messages	—	2	—	8	✅

Baseline: unavailable · PR result: 008162e · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 106.1 KiB
• Claude decoded thread snapshot: 106.4 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

[macroscopeapp]

Approvability

Verdict: Approved at 008162e

Macroscope's review found this PR approvable — This is a focused Codex V2 bug fix that restores execution of explicitly typed skill mentions while preserving ordinary currency amounts and existing dollar syntax. The production change is localized and covered by direct and replay tests for both turn start and steering.

Notes:

• No code objects were reviewed. Approvability was decided on eligibility alone.

^{You can add or adjust custom eligibility rules. Learn more.}

Dismissing prior approval to re-evaluate 008162e