The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes Architecture Decision Records (ADRs) and audit reports which may contain untrusted content. This content is interpolated into briefs for validation subagents like solutions-architect or security-engineer.
Ingestion points: The skill reads .aidoc/audit/05_ADR-audit.md and various ADR files from the workspace (SKILL.md).
Boundary markers: Instructions do not define specific delimiters or instructions to prevent validation subagents from interpreting ingested document content as instructions.
Capability inventory: The skill performs file system writes, creates backups in tmp/backup/, and dispatches multiple subagent types to validate applied fixes.
Sanitization: No explicit sanitization or filtering of the ingested ADR content is mentioned before it is passed to the subagent validation loop.
[COMMAND_EXECUTION]: The skill executes local bash commands to manage workflow timing and session state.
Evidence: Uses Bash: date +%s and Bash: mkdir -p to initialize tracking files (.skill-start.fixer) and check for soft deadlines (1500s) to handle timeouts.
[DYNAMIC_EXECUTION]: The skill dynamically selects and dispatches subagents based on finding personas.
Evidence: Maps lenses such as architect, tech_lead, and chaos_engineer to specific subagent types (e.g., aidoc-flow:solutions-architect) at runtime for parallel validation tasks.

doc-adr-fixer