The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses Bash, Grep, and Glob to perform consistency checks across local project directories including knowledge/, tracks/, and memory/ blocks.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user counter-arguments and converts them into standing rules, creating a vector for persistent instruction poisoning.
Ingestion points: Processes user input triggered by phrases like "is that right?", "something seems off", and "what's your basis?".
Boundary markers: Implements an "Evidence gate" in Step 1 to prevent arbitrary overrides and a mandatory "User Gate Required" diff review in Step 4.5.
Capability inventory: Authorized to use high-privilege file manipulation tools (Write, Edit) and shell access (Bash).
Sanitization: Relies on the user manually reviewing a git diff output to verify changes before they are committed.
[DYNAMIC_EXECUTION]: The skill performs self-modification by updating its own SKILL.md file (Validation Ledger) at runtime based on the execution history of the session.
[PERSISTENCE]: The skill creates and modifies persistent files like CLAUDE.md, memory/*.md, and knowledge/ assets to establish behavior-altering rules that persist across future AI sessions.

verify-bidirectional