Purpose

Systematically evaluate tickets (issues, tasks, stories) for completeness, clarity, and actionability. Check required fields, validate acceptance criteria, score scope definition, identify missing dependencies, sanity-check effort estimates, and flag ambiguous language that would block an implementer from starting work.

When to use

• A batch of tickets has been created and needs quality assurance before sprint planning.
• An agent-generated ticket set needs validation before assignment.
• A backlog grooming session requires an objective quality score for each ticket.
• A milestone's tickets need dependency-chain verification.
• Tickets are being imported from external sources and need standardisation checks.

Do NOT use when

• The task is to create new tickets — use

  ticket-creator

  instead.
• The task is to implement what a ticket describes — use an implementer skill.
• The audit scope is code quality, not ticket quality — use

  code-review

  .
• Only a single field needs updating — just edit the ticket directly.

Operating procedure

1. List all ticket files by running

   find . -path '*/tickets/*.md' -o -path '*/issues/*.md' | sort

   or reading the ticket index/manifest file.
2. For each ticket, extract and check these required fields: Title, Description, Acceptance Criteria, Priority, Labels, and Assignee (or "unassigned").
3. Score each ticket on a 1–5 scale for Clarity (can an implementer start without asking questions?), Scope (is the boundary well-defined?), and Testability (can completion be objectively verified?).
4. Validate acceptance criteria: each criterion must be a concrete, verifiable statement — flag any that use words like "should work well", "appropriate", or "as needed" as vague.
5. Check for dependency completeness: if a ticket references another ticket, verify the referenced ticket exists and is not in a contradictory state (e.g., blocked ticket depending on another blocked ticket).
6. Verify effort estimates: if present, check that they are consistent with scope (flag tickets estimated at 1 point but spanning 5+ files, or 8 points for a single config change).
7. Run

   grep -l 'TODO\|FIXME\|TBD\|PLACEHOLDER' <ticket-files>

   to find tickets with unresolved placeholders.
8. Check for duplicates: compare ticket titles and descriptions pairwise and flag any with >80% textual overlap.
9. Compile a summary table:

   | Ticket ID | Title | Clarity | Scope | Testability | Issues Found | Status |

   .
10. Produce a prioritised list of tickets that need revision before they can be assigned.

Decision rules

• A ticket missing acceptance criteria always scores Testability ≤ 2.
• A ticket with no clear scope boundary (affects "multiple areas" without listing them) scores Scope ≤ 2.
• Duplicate tickets: keep the more detailed one and recommend closing the other.
• If a ticket has 0 issues, mark it READY. If 1–2 minor issues, mark NEEDS_POLISH. If 3+ issues, mark REWRITE.
• Never rewrite tickets yourself — flag them with specific fix instructions.

Output requirements

1. Audit Summary Table — one row per ticket with scores and status.
2. Issue Detail List — for each flagged ticket: ticket ID, issue type, quoted problematic text, and specific fix instruction.
3. Dependency Graph Check — list of broken or circular dependency chains.
4. Duplicate Report — pairs of tickets with high overlap and recommended action.
5. Overall Backlog Health — percentage of tickets that are READY, NEEDS_POLISH, and REWRITE.
6. Top 5 Fixes — the five highest-impact improvements to make first.

References

• Project ticket directory or manifest file
• Ticket templates if defined in

  .github/ISSUE_TEMPLATE/

  or

  docs/templates/

• Definition of Done document if present in the repo

Related skills

• ticket-creator

  — creates tickets that this skill audits

• team-leader

  — consumes audit results to decide assignment readiness

• planner

  — generates ticket sets that benefit from auditing

• backlog-verifier

  — validates backlog state at a higher level

Failure handling

• If no ticket files are found, search for GitHub Issues via the API or report "no tickets found in repo."
• If a ticket uses a non-standard format, attempt best-effort field extraction and note format deviations.
• If the ticket set exceeds 50 tickets, audit in batches of 15 and produce incremental reports.
• If scoring is ambiguous for a ticket, assign the lower score and add a note explaining the uncertainty.