Revert "fix(ai-review): default opus-model to the bare "opus" alias"

[Hicham Zinalabdin (hichamza)]

Reverts #47.

Summary

• Restores opus-model's default to the pinned claude-opus-5 in both the action input default and the routing step's shell fallback, and reverts the matching README table entry.

Test plan

• node --test lib/*.test.js — 145/145 passing

✅ ai-qa: PASS

Deploy health: healthy · findings P0 0 · P1 0 · P2 0 · P3 0

Last updated by ai-qa · QA run

🤖 AI Review — checklist verification

• ❔ node --test lib/*.test.js — 145/145 passing — Test execution is out of scope for this review session (no toolchain access; test_execution reported as skipped). Grepped ai-review/lib/*.test.js for claude-opus-5 and opus-model — no matches, confirming the two changed lines aren't covered by these tests, which supports the 'unaffected' framing. The raw 145/145 pass count itself could not be independently reproduced without running the suite.

Verification evidence:

• git rev-parse HEAD~1 HEAD → 60a91e6 / 2bc9e67 — exactly matches manifest.json's base_sha/head_sha
• git diff 60a91e6..2bc9e67 -- ai-review/README.md ai-review/action.yml (compared against PR #47's own diff: git diff 1035181..d000920 -- .) → PR Revert "fix(ai-review): default opus-model to the bare "opus" alias" #48's 3 hunks are the exact line-for-line inverse of PR fix(ai-review): default opus-model to the bare "opus" alias #47's 3 hunks (README table cell, action.yml input default, shell fallback ${OPUS:-...}) — no additional or missing changes
• Grep 'opus-model|OPUS:-opus|default.*opus' across the repository → Only ai-review/action.yml, ai-review/README.md, docs/adr/0003-....md, and docs/plan.md match; the two docs files reference 'opus-model' only in prose about routing behavior, not the specific default string, so neither goes stale from this revert
• Grep 'opus-model' in .github/workflows/ → No matches
• grep 'claude-opus-5|opus-model' ai-review/lib/*.test.js → No matches

Last updated by ai-review · verdict: pass.

[mtm-bot]

✅ ai-qa: PASS

Merge commit 4d6ea24a4702

Signal	Result
Deploy health (https://raw.githubusercontent.com/EdulyCom/github-actions/main/README.md)	✅ healthy
QA review	⚠️ did not run — reporting deploy-health signal only

Posted by mtm-bot[bot] · workflow run

[mtm-bot]

✅ PASS

0 P2 / 1 P3 finding(s) noted — non-blocking.

Confidence: 100 · Merge risk: low
P0: 0 · P1: 0 · P2: 0 · P3: 1

P0 — Blockers

None.

P1 — Should Fix

None.

P2 — Nice-to-Have

None.

P3 — Nits

• PR title Revert "fix(ai-review): default opus-model to the bare "opus" alias" doesn't carry a Conventional Commits type prefix (manifest flags title_ok: false). This is the standard git-generated revert title, and no commitlint/title-lint job exists in .github/workflows/ (checked: actionlint, ai-qa-selftest, parity, selftest, unit, zizmor), so nothing in CI actually enforces this — noted only because the rubric surfaces it.

Strengths

• Exact, complete revert. Diffed this PR's change (60a91e6..2bc9e67) against PR #47's own change (1035181..d000920) and confirmed they are the precise line-for-line inverse across all three touched spots: the README input table row, the action.yml input's default:, and the shell routing step's ${OPUS:-...} fallback. Nothing extra, nothing missing.
• No orphaned references left behind. Repo-wide grep for opus-model / OPUS:-opus / the bare-alias default found only the two changed files plus two docs (docs/adr/0003-...md, docs/plan.md) that discuss routing behavior in prose without citing the specific default string — neither needs updating for this revert to be internally consistent.
• No blast radius beyond the default itself. No consumer workflow in this repo sets opus-model explicitly, so the revert only changes what a caller gets when it omits the input — no call site breaks.
• Test-quality claim holds up structurally, even though it can't be executed here: grepping ai-review/lib/*.test.js for claude-opus-5/opus-model turns up nothing, consistent with the PR's "unaffected; these two lines aren't covered by unit tests" note. This is a pure composite-action-wiring/docs change with no unit-testable logic, so no missing-tests penalty applies.