The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill analyzes external project data which could potentially contain malicious instructions, though it is used here for legitimate auditing. Ingestion points: Git logs (git log --grep), CHANGELOG files, Architecture Decision Records (ADRs), and project memory entries are used to verify failure signatures (SKILL.md Step 1). Boundary markers: The instructions do not specify the use of delimiters or 'ignore' instructions when processing these external files. Capability inventory: The skill utilizes file reading (grep), file writing (creating rules in .claude/rules/), and shell execution for countermeasure verification (SKILL.md Step 3a, 4). Sanitization: The skill includes a specific security guideline to avoid storing secrets or sensitive conversation transcripts in durable records (references/failure-analysis.md).
[DYNAMIC_EXECUTION]: The skill guides the agent to generate and execute local scripts as part of a verification workflow for technical safeguards. Evidence: Instructions include generating a shell script (hook) and executing it with 'bash hook.sh' to verify that it correctly blocks prohibited actions (SKILL.md Step 4).

recurrence-prevention