From engineer-agent
Compare pipeline artifacts across a boundary (spec↔design or design↔tickets) and produce a checklist of gaps. Use this skill when processing a gap-audit queue item.
How this skill is triggered — by the user, by Claude, or both
Slash command
/engineer-agent:audit-gapssonnetThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Compare two pipeline artifacts and produce a checklist of gaps, mismatches, and ambiguities.
Compare two pipeline artifacts and produce a checklist of gaps, mismatches, and ambiguities.
Read, Write — read/write queue itemsGrep, Glob — search completed queue items for cross-referencesmcp__slite__get-note — re-fetch Slite content if neededmcp__atlassian__getJiraIssue — re-fetch Jira ticket details if neededBash — run gh CLI commands for GitHub IssuesA queue item file in ~/.claude/engineer-agent/queue/incoming/ with type: gap-audit. The frontmatter includes a boundary field (spec-design or design-tickets). The ## Context section contains the full content of both artifacts being compared, structured as ### Left Artifact and ### Right Artifact.
Read the queue item file. Extract:
boundary from frontmatter (spec-design or design-tickets)### Left Artifact under ## Context### Right Artifact under ## ContextFor spec-design boundary:
From the spec (left), extract:
From the design doc (right), extract:
For design-tickets boundary:
From the design doc (left), extract:
From the tickets (right), extract:
Perform bidirectional comparison:
For spec-design boundary:
For design-tickets boundary:
For every mismatch found, classify it as one of:
missing-right — present in left artifact only, missing from right. E.g., a spec requirement with no design section, or a design component with no ticket.missing-left — present in right artifact only, missing from left. E.g., a design section that doesn't trace to any spec requirement, or a ticket with no design component.diverged — both sides address it but they disagree on scope, approach, or details.ambiguous — unclear mapping between the two sides, needs human judgment to determine if there's a real gap.For each gap, generate a suggested fix where possible:
Write the ## Draft Response section in this format:
## Draft Response
### Gap Audit: {Spec vs Design Doc | Design Doc vs Tickets}
**Left:** {artifact title} ({url})
**Right:** {artifact title} ({url or "tickets"})
**Gaps Found:** {N}
### Checklist
#### 1. {Gap title}
- **Type:** missing-right | missing-left | diverged | ambiguous
- **Left side:** {what the left artifact says, or "Not present"}
- **Right side:** {what the right artifact says, or "Not present"}
- **Suggested action:** {neutral description of what to do}
- **Draft:** _(if applicable)_
{draft content — ticket body, section outline, requirement text, or reconciled version}
#### 2. {Gap title}
- **Type:** missing-right | missing-left | diverged | ambiguous
- **Left side:** {what the left artifact says, or "Not present"}
- **Right side:** {what the right artifact says, or "Not present"}
- **Suggested action:** {neutral description of what to do}
- **Draft:** _(if applicable)_
{draft content}
{repeat for each gap found}
### Summary
- {N} items only in {left artifact type} (missing from {right artifact type})
- {N} items only in {right artifact type} (missing from {left artifact type})
- {N} items present in both but diverged
- {N} ambiguous mappings needing human judgment
If no gaps are found for a boundary, write:
## Draft Response
### Gap Audit: {boundary description}
**Left:** {artifact title} ({url})
**Right:** {artifact title} ({url})
**Gaps Found:** 0
No gaps detected. The artifacts appear well-aligned across this boundary.
Update the queue item's frontmatter status to drafted and move it from ~/.claude/engineer-agent/queue/incoming/ to ~/.claude/engineer-agent/queue/drafts/ (write to new location, delete from old).
Report: "Gap audit drafted for {boundary} boundary: {N} gaps found. Review with /engineer-agent review-queue gap."
Creates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.
npx claudepluginhub tomharris/engineer-agent --plugin engineer-agent