[qa-lead S5] Reconcile personas/qa-lead/README.md with the shipped runtime #1648

Story

As someone told by persona-standards §7 that qa-lead is the worked example to copy,
I want personas/qa-lead/README.md to describe the persona that actually shipped,
so that the reference implementation stops teaching a design that was superseded.

The problem, verified on origin/main (85658ac)

The README's "Status: draft" section makes three claims that are false:

And promotion step 1 is actively misleading:

1. Wire a dedicated advisory workflow (e.g. mention- or check_run-triggered test review) as a caller stub + reusable, and add its runtime: block.

persona-standards §4.1 says the opposite: "ONE router serves every persona … a stub per agent is the drift the manifest exists to prevent." That is also what shipped — persona-runner-reusable.yml is explicitly "ONE runner for EVERY persona — manifest/convention driven, not one workflow per persona." A reader following this README would build the thing the standard forbids.

The README has been stale since #1293 merged the runtime (2026-07-18).

Acceptance Criteria

1. The three false claims are corrected. The README describes the shipped architecture: one shared mention router (petry-projects/.github, persona-mention-reusable.yml) → repository_dispatch: persona-mention → one shared runner (persona-runner-reusable.yml) → the persona's prompts/<id>/advisory.md.
2. Promotion step 1 is rewritten to match §4.1 — a new persona adds a manifest and an advisory prompt; it does not add a workflow. Anyone following the steps must end up with something the standard permits.
3. The read/write split is documented, because it is the security property the whole framework rests on: the agent runs with github.token and cannot post; the workflow enforces the <!-- persona:<id> --> marker and posts with the PAT. Cite feat: implement issue #846 — [Phase 2] Cache the LSP index and enforce the 30s cold-start SLA with auto-skip #860 (prompt-only marker enforcement → 1,481 acks in 4.5h) so the next reader understands why it is built that way.
4. prompts/qa-lead/advisory.md is documented as part of the persona — it is the actual behavior surface and the README currently never mentions it.
5. The remaining accurate content is preserved: the role-not-person rationale (§1.6), the team-handle-not-user-account rationale, the vendored-not-forked posture, and the frameworks/ no-hand-edits rule.
6. The status section states what is genuinely still outstanding — no agents.qa-lead entry in canary-rings.json, still next-ring only (persona-mention/v1-ring0 / v1-ring1 / v1-stable do not exist), eval gate not yet runnable ([qa-lead S2] Make the qa-lead eval set scorable and capture a first held-out baseline #1645) — and does not restate anything the manifest already says. §4.1's rule is that the manifest is the index-of-record and the docs restate none of it.
7. markdownlint-cli2 clean; internal links resolve.

Tasks / Subtasks

• Rewrite the Status section against the shipped runtime. (AC: test issue from agent #1, feat: feature ideation framework — adversarial proposer/challenger/synthesizer #6)
• Rewrite the promotion steps to match §4.1. (AC: Go-live improvements for PR review agent #2)
• Document the dispatch chain, the read/write split, and the advisory prompt. (AC: Add @claude delegation, auto-merge, and rebase handling #3, Optimize review: small-PR and incremental fast paths #4)
• Verify the preserved sections are still accurate. (AC: feat: add Copilot engine support via REVIEW_ENGINE toggle #5)
• Lint and link-check. (AC: Workflow should skip no-op PRs in MAX_PRS count #7)

Dev Notes

• Docs-only. No manifest, workflow, or prompt behavior changes. If writing this surfaces a substantive gap, file it rather than fixing it here.
• This README is load-bearing beyond this persona: persona-standards §7 points readers at it as the worked example, so its errors propagate into every persona copied from it. Three of the four ops personas ([bmad-onboard S1] Onboard sre-lead persona #1307/[bmad-onboard S2] Onboard devops-lead persona #1308/[bmad-onboard S3] Onboard security-lead persona #1309) were onboarded after it went stale.
• Do not restate the manifest. The temptation is to explain each surface and trigger in prose; §4.1 explicitly forbids the docs from restating per-persona decisions, because two sources drift. Link to persona.yml and interaction.yml instead.
• [qa-lead S3] Serve the pull_request surface so qa-lead runs without being summoned #1646 and [qa-lead S4] Make persona.yml's declared surfaces agree with what is actually deployed #1647 will amend this file again when the surface set changes. Write today's truth; do not try to pre-describe their end state.

Project Structure Notes

personas/qa-lead/README.md only.

References

• petry-projects/.github standards/persona-standards.md §1.6, §4.1, §7
• .github/workflows/persona-runner-reusable.yml, .github/workflows/persona-mention.yml
• prompts/qa-lead/advisory.md, personas/qa-lead/persona.yml, personas/qa-lead/interaction.yml
• The persona runtime — dispatch side of the mention framework #1293 (the runtime that made the README stale), Rename murat → qa-lead and make it @-addressable #1279, Add Murat persona instance — worked example for the persona standard #1174
• docs/postmortems/2026-06-pr-860-runaway.md