Templates¶

All agent outputs follow structured templates validated by bash scripts.

Code profile templates¶

Finding template¶

Every code review finding uses this format:

Finding ID: <TAG>-NNN
Specialist: <specialist name>
Severity: [Critical | Important | Minor]
Confidence: [High | Medium | Low]
File: <file path>
Line: <line number or range>
Title: <max 200 chars>
Evidence: <max 2000 chars - must cite specific code>
Recommended fix: <max 1000 chars>
Source Trust: [First-Party | Third-Party | Generated | Vendored | Test]

Field rules:

Finding ID prefix must match specialist tag (SEC, PERF, QUAL, CORR, ARCH)
Evidence must exceed 100 characters (auto-demoted to Minor otherwise)
Source Trust is required for security findings, optional for others
Recommended fixes are scanned for destructive patterns

Challenge response template¶

Used during Phase 2 cross-agent debate:

Challenge: <finding-id>
Challenger: <tag>
Type: [False Positive | Severity Inflation | Missing Context | Duplicate]
Argument: <max 1000 chars>
Evidence: <file:line citations>

Report template¶

The Phase 4 report includes up to 14 sections:

Executive summary with agreement level
Review metadata (target, specialists, iterations, budget)
Critical findings
Important findings
Minor findings
Escalated disagreements
Dismissed findings with rationale
Challenge round highlights
Co-located findings
Remediation roadmap
Budget summary
Reference modules used
Appendix: raw specialist outputs (optional)
Appendix: convergence data (optional)

Triage templates¶

Triage mode uses specialized templates:

Triage finding: Verdict (Fix/No-Fix/Investigate), confidence, analysis per comment
Triage report: Summary of all verdicts with coverage gaps (when --gap-analysis is used)
Triage input schema: Expected format for comment sources

Other templates¶

Jira template: For Phase 5 ticket drafts
Delta report template: For --delta mode incremental reports
Sanitized document template: Post-sanitization output format

Strategy profile templates¶

Finding template¶

Strategy findings include a verdict instead of file:line references:

Finding ID: <TAG>-NNN
Specialist: <specialist name>
Severity: [Critical | Important | Minor]
Confidence: [High | Medium | Low]
Document: <strategy document name>
Citation: <section, paragraph, or AC reference>
Title: <max 200 chars>
Evidence: <max 2000 chars - must cite specific strategy text>
Recommended fix: <max 1000 chars>
Verdict: [Approve | Revise | Reject]

Challenge response template¶

Same structure as code profile but with document citations instead of file:line.

Report template¶

Strategy reports include per-document verdicts and requirements extraction alongside the standard sections.

Requirements template¶

Extracted requirements from strategy documents, structured for downstream consumption.