The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to perform state recovery and system maintenance using command-line tools. Evidence includes instructions to use git checkout to revert unsuccessful workflow changes and python -m workflowhooker check to manage file and git locks during emergency shutdown sequences.\n- [INDIRECT_PROMPT_INJECTION]: The skill implements the PROTOCOL-ANCHOR-RESET heuristic, which directs the agent to read and follow directives from local files such as CLAUDE.md, AGENTS.md, and START.md as authoritative 'North Star' documents to reset its goal state. This establishes a surface where an attacker capable of modifying these files could override the agent's behavior.\n
Ingestion points: Local configuration and documentation files CLAUDE.md, AGENTS.md, and START.md (referenced in SKILL.md and translations).\n
Boundary markers: Absent; no delimiters or explicit warnings are provided to prevent the agent from obeying malicious instructions embedded within these files.\n
Capability inventory: The skill utilizes shell command execution capabilities via git and python tools.\n
Sanitization: Absent; no procedures for validating or sanitizing the content of these anchor documents are specified before they are processed as authoritative directives.

wayfinding-routing