| name | linear-evidence-review |
| description | Posts idempotent Linear evidence receipts from executable-stories review reports using Linear MCP. Use when the user asks to update Linear from an Evidence Review, post a green-tick receipt to linked Linear issues, summarize verified behaviour on tickets, or use Linear MCP with executable-stories reports. |
Linear Evidence Review
Use this skill for the interactive agent convenience path: an IDE/CLI agent reads an executable-stories review, finds linked Linear issues, and posts or updates a short receipt through Linear MCP.
This is not the CI write-back path. Do not promise headless receipt-on-merge here; Linear MCP usually relies on interactive OAuth client setup. CI automation is a later product/API decision once the receipt workflow is proven stable.
- The executable-stories report is the source of truth.
- Linear issues are provenance and workflow records.
- This skill uses Linear MCP only; do not add executable-stories product code.
- Do not create Confluence/Jira variants yet. Learn the Linear receipt workflow first.
- Do not rename tests after tickets. Behaviour/domain file names preserve review correlation, e.g.
src/checkout/regional-routing.story.test.ts can correlate to src/checkout/regional-routing.ts; LIN-123.test.ts cannot.
Prerequisites
This skill is agent-agnostic: it only needs a Linear MCP server connected in whatever agent you run. The remote endpoint is https://mcp.linear.app/mcp; first use triggers an interactive OAuth login. Once a Linear MCP tool is available, the workflow below is identical across agents.
claude mcp add --transport http linear https://mcp.linear.app/mcp
codex mcp add linear --url https://mcp.linear.app/mcp
For any other MCP-capable client (Cursor, VS Code, Windsurf, Gemini CLI, …), add the remote server https://mcp.linear.app/mcp through that client's MCP configuration and complete the OAuth prompt. The rest of this skill does not depend on which agent is used.
Workflow
- Generate or inspect the Evidence Review output.
- Prefer existing review artifacts if present.
- Otherwise run the project's normal test/report command, then
executable-stories review.
- Read ticket provenance from the canonical report model:
testCase.story.tickets.
- Each ticket is a normalized
{ id, url? } reference.
- Do not invent a new ticket field.
- Identify Linear tickets, then fetch those issues through Linear MCP.
- Treat a ticket as Linear when
ticket.url has host linear.app.
- Do not infer from
LIN-123 style IDs; team keys are not reliable. If a ticket has no URL and is ambiguous, ask.
- If an issue is missing or inaccessible, report it and continue with the rest.
- Build a short receipt.
- Scope each issue receipt to claims whose
testCase.story.tickets include that issue.
- Include verified behaviours, evidence strength, uncovered/weak counts, and the report link or artifact path.
- Mention that the report is authoritative.
- Confirm before writing.
- Show target issues and planned receipt text.
- Proceed only after the user approves, unless the user already explicitly said to post/update Linear.
- Update Linear idempotently.
- Search the issue comments for the marker below.
- If found, update that comment in place.
- If not found, create one comment.
Use one stable marker per issue and PR/head ref:
<!-- executable-stories:evidence-review issue=LIN-123 ref=<branch-or-pr> -->
Use PR number when available, otherwise the branch name. Never use commit SHA: it changes on every push and breaks idempotency. Never create a second receipt with the same marker.
Use ✅ only when all claims linked to that issue passed with moderate or strong evidence and no linked changed source is uncovered. Use ⚠️ if any linked claim failed, has weak/none evidence, or has uncovered changed source.
Receipt Shape
Keep the comment concise:
<!-- executable-stories:evidence-review issue=LIN-123 ref=PR-42 -->
✅ Executable Stories evidence receipt
Verified behaviours:
- Checkout selects the right regional payment route — strong evidence
- Currency fallback keeps the existing checkout total — weak evidence
Authoritative report: <link-or-path>
Note: this Linear issue is provenance. The executable report remains the source of truth.
Do not paste the full report into Linear.
What Not To Do
- Do not copy the whole report into Linear.
- Do not post to issues without showing the plan first, unless the user already explicitly said to post.
- Do not close or move issues unless the user explicitly asks.
- Do not create CI automation or API scripts from this skill.
- Do not create new issues for every warning unless requested.
- Do not treat Linear titles/descriptions as requirements truth.
- Do not use ticket-named test files as a convention.