Agent Skills: Structured Output

Table of Contents

• When to Use
• Activation Patterns
• Required TodoWrite Items
• Step 1: Select Template (structured-output:template-selected)
• Step 2: Format Findings (structured-output:findings-formatted)
• Step 3: Assign Actions (structured-output:actions-assigned)
• Step 4: Attach Appendix (structured-output:appendix-attached)
• Output Quality Checklist
• Exit Criteria

Structured Output

When To Use

• When finalizing any review or analysis.
• To format findings in a consistent way that names specific next steps.
• Before presenting results to stakeholders or committing them to documentation.

When NOT To Use

• Capturing evidence during analysis - use proof-of-work
• Reviewing changes - use diff-analysis or review-core first

Activation Patterns

Trigger Keywords: format, structure, deliverable, report, organize, present, consistent Contextual Cues:

• "format this as a report" or "structure the output"
• "create a deliverable" or "present these findings"
• "organize this consistently" or "standardize the format"
• "make this actionable" or "prepare for stakeholders"

Auto-Load When: Finalizing any analysis deliverable or when consistent formatting is requested.

Required TodoWrite Items

1. structured-output:template-selected
2. structured-output:findings-formatted
3. structured-output:actions-assigned
4. structured-output:appendix-attached

Mark each item complete as you finish the corresponding step.

Step 1: Select Template (structured-output:template-selected)

• Choose output format based on deliverable type:
  • Review Report: Summary, Findings, Recommendations, Evidence.
  • PR Description: Summary, Changes, Test Plan, Notes.
  • Release Notes: Highlights, Breaking Changes, Fixes, Credits.
  • Incident Report: Timeline, Impact, Root Cause, Remediation.
• Confirm audience and required detail level.

Step 2: Format Findings (structured-output:findings-formatted)

• Use consistent finding structure:

  ### [SEVERITY] Finding Title
  **Location**: file.rs:123
  **Anchor**: `verbatim source text copied from line 123`
  **Category**: Security | Performance | Correctness | Style
  **Description**: Brief explanation of the issue.
  **Evidence**: [E1, E2] - Reference to evidence log.
  **Recommendation**: Specific remediation steps.
  

  Verification: Run the command with --help flag to verify availability.
• Severity levels: CRITICAL, HIGH, MEDIUM, LOW, INFO.
• Order findings by severity, then by file location.
• Anchor is mandatory and grounds the finding. Copy the exact source text at Location (not a paraphrase). It is what a second pass re-reads to confirm the finding is real. A finding whose anchor does not appear at its cited line is treated as a hallucination and dropped. The check is mechanical: imbue:review-core Step 6 runs plugins/imbue/scripts/citation_verifier.py over the findings.

Step 3: Assign Actions (structured-output:actions-assigned)

• Convert findings to action items with assignee and priority:

  ## Action Items
  - [ ] [HIGH] Fix SQL injection in auth.py:45 (@security-team, P1)
  - [ ] [MEDIUM] Add input validation to API endpoint (@backend, P2)
  - [ ] [LOW] Update deprecated dependency (@devops, P3)
  

  Verification: Run the command with --help flag to verify availability.
• Include owner assignment where known.
• Add priority indicators (P1/P2/P3) for triage.
• Note dependencies between actions.

Step 4: Attach Appendix (structured-output:appendix-attached)

• Compile supporting materials:

  ## Appendix
  ### A. Commands Run
  [Full evidence log from imbue:proof-of-work]
  
  ### B. External References
  [Citations and documentation links]
  
  ### C. Raw Data
  [Large outputs, full diffs, or data exports]
  

  Verification: Run the command with --help flag to verify availability.
• Keep main report concise; details in appendix.
• validate appendix is navigable with clear section headers.

Output Quality Checklist

Before finalizing:

• [ ] Every finding carries a verbatim Anchor the citation verifier resolved (no unverified findings ship).
• [ ] All findings have evidence references.
• [ ] Severity levels are justified.
• [ ] Recommendations are specific and name the next step.
• [ ] No orphaned sections or placeholder text.
• [ ] Format renders correctly in target medium (GitHub, Confluence, etc.).

Exit Criteria

• Todos completed with formatted deliverable.
• Output follows selected template structure.
• Stakeholders can act on findings without clarification.