Acceptance Criteria Doc

Acceptance Criteria Doc Generator

Produce three things for a given Jira epic:

1. A live Confluence page titled <EPIC-KEY> - ACs in the TL1 space, inside the Assets - Acceptance Criteria folder (parentId: 401775165477), on cloud consensyssoftware.atlassian.net.
2. A Jira comment on the epic linking the doc.
3. The AC doc link appended to every child issue's description (stories, tasks, spikes — all children).

Example outputs: ASSETS-3647 - ACs (Market Insights), ASSETS-3645 - ACs (CTA Optimisation).

Inputs

Required: the epic (key or URL). Everything else is optional and should be mined rather than requested:

• Epic description & remote links — often contain the feature flag name, Definition of Done link, PRD, ADR/tech-spec, Figma/Replit links, and a "source epic" (e.g. the Mobile epic a parity feature is ported from). Extract all of these.
• Repositories — if the user names repos, investigate them; if the feature is a port/parity of an existing implementation, the source repo is the primary investigation area even if unnamed.
• PRD / Figma / Replit / ADR — link them in the Documentation table when available. Don't block on missing ones; note the gap to the user in your final summary instead of padding the doc.

Workflow

1. Gather Jira context

• getJiraIssue on the epic (markdown format) — read the description carefully for flags, links, and scope checklists.
• searchJiraIssuesUsingJql with parent = <EPIC-KEY> ORDER BY created ASC to get the child issues. Each testable story becomes a criteria group; spikes/discovery tasks get the AC link but not a criteria row (they aren't testable).
• If the epic references a source epic (ported feature), fetch its children too — their titles reveal the real scope (analytics, e2e tests, edge cases) that the new epic's stories may summarize away. Large JQL responses get saved to a file; extract just key/status/summary with jq rather than reading the whole payload.

2. Fetch the structural reference

The house style lives in the previous AC docs, not in this skill. Call getConfluencePageDescendants on the folder (pageId: 401775165477) and pick the most recently created <KEY> - ACs page, then fetch it with getConfluencePage (contentFormat: "html"). Mirror its structure, heading order, table layouts, and cell markup exactly — this keeps every new doc consistent with whatever the team's current format is, even as it evolves. When reusing its markup as a scaffold, strip all data-local-id attributes (Confluence assigns fresh ones).

3. Investigate repositories

In parallel with the above, spawn a general-purpose subagent to research the reference implementation — this is what makes the criteria concrete instead of generic. Have it use gh (PR search, code search) rather than cloning. Ask it to report, with exact names and file paths:

• Merged PRs (number, title, URL, one-liner) that built the feature
• Component names/paths
• Feature flag names (both the new surface's flag and the reference implementation's)
• Analytics event names and their properties, and where they fire
• Data source (API/controller/package)
• Empty/error/loading state handling worth testing

4. Compose the document

The structure (matching the reference doc from step 2) is:

1. h1 — "Acceptance Criteria for "
2. h1 Overview — one paragraph: what the epic delivers, referencing the source implementation if it's a port.
3. h2 Documentation — two-column table (Document Type | Link): epic, Definition of Done, source epic, repo(s), data source, feature flag, plus PRD/Figma/ADR when available.
4. h1 Testing Scenarios — intro sentence ("must be tested successfully prior to feature release…"), then the criteria table with columns: Criteria (500) | Expected Result (220) | one 90-wide checkbox column per platform.

Choosing QA sign-off columns: infer from the epic's scope. A parity/port epic gets a column per platform (e.g. Extension + Mobile); a single-platform epic gets one column. If genuinely ambiguous, ask the user before creating the page.

Writing the criteria — the audience is devs, PMs, EMs, and QA skimming during sign-off. Criteria that are long or clever get ignored, so:

• Group rows by story. First row of each group starts with a bold group title linking the story (e.g. **Market Insights entry card** — see ASSETS-3682). Follow-up rows in the group have no title.
• One behavior per row, written as three Given/When/Then bullets. If a Then clause needs an "and… and… and…", split the row.
• Expected Result is one plain sentence — the thing QA verifies, no restating the Given/When.
• Cite evidence inline: reference PRs, exact event names, flag names, and property names in code formatting. Concrete names are what let QA actually verify parity.
• Always include a feature-flag gating row (flag off → no UI, no events) and a performance row (no regression vs. current experience) at the end.
• Target 8–12 rows total. If the epic is bigger than that, tighten the Expected Results before adding rows; a doc nobody reads verifies nothing.

5. Create the page and link everywhere

1. createConfluencePage with spaceId: "TL1", parentId: "401775165477", subtype: "live", contentFormat: "html", title <EPIC-KEY> - ACs.
2. Take the tiny link from the response (_links.tinyui, e.g. /x/GoC_i10 → https://consensyssoftware.atlassian.net/wiki/x/GoC_i10) and use it in all Jira writes.
3. addCommentToJiraIssue on the epic (contentFormat: "markdown"): link the doc and one sentence on what it covers, linking the stories.
4. For each child issue: append **Acceptance Criteria:** [<EPIC-KEY> - ACs](<tiny link>) to its description via editJiraIssue — but only after fetching the current description in-context, so existing text is preserved verbatim (and so the write isn't blocked as a blind overwrite; see Gotchas).

6. Report

End with: the doc link, the epic comment confirmation, which issues were updated, and anything discovered in the repo investigation that the epic's stories don't cover (candidate out-of-scope items). That last part regularly surfaces real scope gaps — don't skip it.

Gotchas (all hit in practice)

• editJiraIssue responses can echo the wrong issue. After batch description updates, verify with one searchJiraIssuesUsingJql read of the edited keys instead of trusting the write responses.
• Blind description writes get blocked. The permission layer rejects editJiraIssue description writes when the current description wasn't visibly fetched first. Fetch, then write fetched-text + appended link.
• Checkbox cells must be exactly <ul data-type="task-list"><li data-type="task-item"><input type="checkbox"> </li></ul> — anything else silently fails to become a Confluence task. Verify this matches the reference doc fetched in step 2.
• Never invent data-local-id attributes; omit them on new content and strip them from copied scaffold markup.
• Don't wrap the body in <html>/<head>/<body>, and keep Given/When/Then bullets as plain <li><p>…</p></li> (list items can't contain headings/tables/panels).

Reusing this for another team

Everything above is team-agnostic except three values: the space key (TL1), the AC folder parentId (401775165477), and the example links. Change those to your team's equivalents and the rest applies unchanged.

Written up for humans at Skill - Acceptance Criteria Creation, which is the page to point teammates at.