agentleFS
geekxxgeekxx/ai-assistant-starter-kit/.claude/skills/sweep-jira/SKILL.md

sweep-jira

Skill1 starsChanged 4 months agoMITView on GitHub

On-demand Jira status sweep for one or all Forward Deploy firms. Reads Story/Task changes under a firm's Epic (scoped by Component) since the last sweep and produces a plain read-only summary; if evidence from elsewhere in a sweep (a transcript, Slack, email) suggests a specific issue's status should change, proposes it for Jeff's approval — it never transitions or edits Jira itself. Trigger on "sweep jira," "jira status," "what's moving in jira for [firm]," "jira sweep.

The file

Show its rules and text

We split this skill into the rules it gives an agent

14 rules. About 0 are common, 6 are distinctive, and 8 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈6

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 8

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Escalate to Vantage (claude-opus-4-8): only when a PROPOSED-CHANGE requires real judgment — e.g., reconciling what a transcript said …

    line 37
  2. Very specific:

    Default scope is all 5 firms (per _shared/fd-firms.md), each swept independently.

    line 45
  3. Distinctive:

    Default is since the last Jira sweep for this firm:

    line 54
  4. Very specific:

    Look in owners_inbox/ (and owners_inbox/archive/) for the most recent file matching jira-sweep_<YYYY-MM-DD>_<firm>.md.

    line 56
  5. Very specific:

    Fallback: if no prior sweep exists for that firm, use the previous 7 days (matches the weekly cadence of fd-blocker-sweep, which this lane …

    line 59
  6. Distinctive:

    If Jeff named a window, that overrides.

    line 61
  7. Very specific:

    Write to owners_inbox/jira-sweep_<YYYY-MM-DD>_<firm>.md (one file per firm.

    line 102
  8. Very specific:

    If the calling context (e.g.

    line 108
  9. Distinctive:

    Never batch-approve; each proposed change needs its own explicit yes.

    line 122
  10. Very specific:

    Every jira-sweep_<date>_<firm>.md ends with:

    line 126
  11. Distinctive:

    Never batch multiple proposed changes into one approval — each is its own yes/no.

    line 142
  12. Distinctive:

    Never treat a sparse firm (few/no Stories) as an error — it is the expected state for CLA, Baker Tilly, and Archer Lewis today.

    line 143
  13. Distinctive:

    Never invent a status change without a cited source (transcript, Slack, email, or Jeff's own statement).

    line 145
  14. Very specific:

    Never confuse the IESFD project with the unrelated SOR-## prefix used by a different team (CUPA) — see _shared/fd-sot-sources.md.

    line 147

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md
<!-- audience: internal -->

Sweep Jira — On-Demand

Reads what changed in Jira for a Forward Deploy firm (or all of them) since the last sweep and reports it plainly. This is a read lane: it never writes to Jira. If the sweep run surfaces evidence that an issue's status should change (from a transcript, Slack message, or email passed in by a calling skill), it produces a PROPOSED-CHANGE block for Jeff to approve — Sol never transitions or edits an issue on its own say-so.

Reads _shared/fd-firms.md for the current firm list and _shared/fd-sot-sources.md for the Jira project key and Epic/Story/Task/Component structure — read both before running; do not hardcode the firm list or project key here.

When to run

  • Jeff says "sweep jira," "jira status," "what's moving in jira for [firm]," or "jira sweep."
  • As part of /full-sweep (Step 4), invoked once per firm, dispatched in parallel across all 5 firms.

Who does the synthesis — Sol, not a subagent (by default)

Unlike fd-blocker-sweep (which reconciles a messy corpus of recaps, Jira, and an open-decisions log into a prioritized, aged list — real synthesis that earns Vantage + claude-opus-4-8), a single-firm JQL query filtered by Component and an update-since date is a small, structured read: pull the issues, group by Story, note status changes. That does not need a subagent.

Default: Sol runs the JQL query directly and writes the summary itself (no specialist dispatch, no model-tier spend beyond Sol's own).

Escalate to Vantage (claude-opus-4-8) only when a PROPOSED-CHANGE requires real judgment — e.g., reconciling what a transcript said against what Jira currently shows, when they disagree, or when more than one issue's evidence needs to be weighed together to decide what to propose. That is bounded judgment over messy input, matching the tier table in CLAUDE.md ("Choosing the Right LLM"). A plain "here's what changed in Jira this week" summary never needs Vantage.

Step 1 — Resolve scope

Default scope is all 5 firms (per _shared/fd-firms.md), each swept independently. If Jeff names one firm, scope to it. When sweeping "all," dispatch the per-firm read in parallel — this is I/O-bound (separate JQL queries), not reasoning-bound, so parallel Sol-driven queries are fine without spinning up 5 subagents.

State the resolved scope in one line and proceed.

Step 2 — Resolve the time window

Default is since the last Jira sweep for this firm:

  1. Look in owners_inbox/ (and owners_inbox/archive/) for the most recent file matching jira-sweep_<YYYY-MM-DD>_<firm>.md. If found, the window starts at that file's date and runs to now.
  2. Fallback: if no prior sweep exists for that firm, use the previous 7 days (matches the weekly cadence of fd-blocker-sweep, which this lane feeds).
  3. If Jeff named a window, that overrides.

State the resolved window per firm in one line (e.g. "Aprio: since 2026-07-28 → now").

Step 3 — Query Jira

Load the tools:

ToolSearch select:mcp__claude_ai_Atlassian__searchJiraIssuesUsingJql,mcp__claude_ai_Atlassian__getJiraIssue,mcp__claude_ai_Atlassian__getVisibleJiraProjects

Resolve the cloudId once via getVisibleJiraProjects or getAccessibleAtlassianResources if not already known (per _shared/fd-sot-sources.md: intuit-prod.atlassian.net).

For each firm in scope, run searchJiraIssuesUsingJql scoped by the firm's Component and the resolved window, e.g.:

project = IESFD AND component = "<Firm>" AND updated >= "<window-start>" ORDER BY updated DESC

Request fields: summary, status, issuetype, priority, assignee, updated, parent. Also fetch the firm's Epic (one getJiraIssue call, or include issuetype = Epic in the same JQL) so the summary can state which Epic each Story/Task rolls up under.

A firm with zero or few Stories (CLA, Baker Tilly, Archer Lewis today) is expected, not an error — report "no Jira activity this window" plainly rather than treating it as a failure.

Step 4 — Produce the summary (read-only, AUTO, no approval needed)

Per firm, group results by Story (= agent) and list:

  • New Stories (created in the window) — new agents entering Jira.
  • Task status changes under each Story — what moved, from what status to what.
  • Anything stuck — an issue whose updated timestamp is old relative to its status (e.g. "In Progress" but untouched for the whole window) — flag, don't diagnose why.
  • The firm's Epic name/key for context.

This part is read-only and delivered automatically — no Jeff approval needed to write the summary, same as fd-blocker-sweep's read-only reporting.

Write to owners_inbox/jira-sweep_<YYYY-MM-DD>_<firm>.md (one file per firm; when running "all 5," produce 5 files, not one combined file, so the per-firm window-resolution in Step 2 keeps working next time). Each file ends with the trace footer (Step 6).

Step 5 — Proposed changes (if evidence warrants — PROPOSE, do NOT write)

If the calling context (e.g. full-sweep passing in transcript/Slack/email findings, or Jeff directly telling Sol something changed) suggests a specific Jira issue's status should change, produce a PROPOSED-CHANGE block — one entry per issue, matching the diff-style convention already used by fd-blocker-sweep's Gate procedure and process-meetings Step 4:

PROPOSED-CHANGE — <ISSUE-KEY> "<summary>"
  Current status:  <status>
  Proposed status: <status>
  Evidence: <source — transcript name/date, Slack message, or Jeff's own statement>

Show this to Jeff. On his explicit OK, Sol (not Vantage, not this skill) may call transitionJiraIssue / editJiraIssue for that specific issue — see the permission note below. Never batch-approve; each proposed change needs its own explicit yes.

Step 6 — Trace footer

Every jira-sweep_<date>_<firm>.md ends with:

---
**Sources:** [JQL query + window + cloudId, per firm]
**Claims marked [Unverified]:** [list, or "none"]

Jira data is itself the source of record for its own fields (status, assignee, dates) — so most lines need no [Unverified] tag. Tag anything inferred beyond what the issue fields state (e.g. "this looks stuck because..." is an inference; the raw updated timestamp is not).

What this skill must NOT do

  • Never call transitionJiraIssue or editJiraIssue without Jeff's explicit approval in that run. This skill only reads Jira and proposes; it does not write, ever, on its own initiative.
  • Never batch multiple proposed changes into one approval — each is its own yes/no.
  • Never treat a sparse firm (few/no Stories) as an error — it is the expected state for CLA, Baker Tilly, and Archer Lewis today.
  • Never invent a status change without a cited source (transcript, Slack, email, or Jeff's own statement).
  • Never confuse the IESFD project with the unrelated SOR-## prefix used by a different team (CUPA) — see _shared/fd-sot-sources.md.

Permissions note

searchJiraIssuesUsingJql, getJiraIssue, and getVisibleJiraProjects are read tools and are allow-listed for autonomous use (see .claude/settings.json). editJiraIssue and transitionJiraIssue are intentionally not allow-listed — invoking either prompts Jeff for approval at the tool-call level, in addition to the propose-then-approve pattern in Step 5. Both gates exist; neither substitutes for the other.

Related: _shared/fd-firms.md, _shared/fd-sot-sources.md, fd-blocker-sweep/SKILL.md (the deeper weekly synthesis this lane feeds), full-sweep/SKILL.md (the orchestrator that invokes this for all 5 firms).

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.