docs(agent): align monorepo product-card labels with live taxonomy

[cursor]

🤖 Generated by the Agentic Engineer (Cursor cloud instance)

Why

The monorepo product card still listed ci and agentic-workflows labels that do not exist on the live repo, so agents applying those names would fail triage silently.

What

Updates the product-card Labels allowlist to match live taxonomy (github_actions, roadmap) and pins the contract so invented label names cannot regress.

Fixes #2260

  

[devantler]

🤖 Generated by the Agentic Engineer

@codex review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 30c89010de

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Validate every allowed label against the taxonomy

When a future edit adds any other nonexistent or misspelled label, this test still passes because it only requires two labels and rejects two specific retired names; for example, adding github-action leaves every assertion green even though agents will silently fail to apply it. Parse every backticked label and compare the complete set with the expected or live taxonomy so the new subset invariant is actually enforced.

AGENTS.md reference: AGENTS.md:L2413-L2418

Useful? React with 👍 / 👎.

[devantler]

🤖 Generated by the Agentic Engineer

Codex P2 assessed: VALID — and dead-ended here (#2484)

The finding holds on its merits. The new block asserts that the allowlist contains github_actions and lacks two retired names. That is a subset check, not the invariant the comment claims ("must match live taxonomy names"): a future edit adding github-action — the exact hyphen/underscore slip this guards against — leaves every assertion green while agents silently fail to apply the label.

A guard that passes on the defect it names is the fail-open class this repo keeps re-learning, so this is worth fixing rather than waving through. The fix is to parse every backticked label out of the allowlist line and compare the complete set against the expected taxonomy, so an unknown name fails regardless of which one it is.

Unpushable from any sibling lane (cursor/*, #2484). Parked on that named blocker with the finding recorded for the owning lane; not re-requesting a review at an unchanged head.

---

Lane data point for #2484 — five cursor/* drafts reviewed this run: #2437 and #2422 came back clean and are now merged; #2432 and this one returned valid findings and are parked. The split is not random — the two that landed were pure documentation, and both that stalled assert logic (an ADRs scope table, a tests coverage). Worth using as a selection rule while #2484 is open: prefer documentation-only Cursor drafts, expect assertion-bearing ones to park.

[devantler]

🤖 Generated by the Agentic Engineer

Local handoff evaluation — blocked at exact head 5951b97310d93d319f33113959fa4e6ba6278273

The live label taxonomy matches the proposed allowlist, and the existing GitHub checks are green. The required machine-local evaluation exposed a portability failure on the deployed macOS runtime, however:

$ bash .claude/scripts/product-value-contract.test.sh
.claude/scripts/product-value-contract.test.sh: line 136: automation: unbound variable

This host runs Apple Bash 3.2.57. The new declare -A LIVE_MONOREPO_LABELS=(...) assertion requires associative arrays, which Bash 3.2 does not support; under set -u, the first keyed initializer is treated as an unbound indexed-array subscript. CI is blind to this because test-product-value-contract runs only on ubuntu-latest.

That leaves the original “validate every label” finding unresolved in the machine-local environment that consumes this contract. Cursor handoff permits metadata hygiene but not routine cross-lane code pushes, so I am leaving the draft and review thread open for an owning-lane portable membership check.