The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses shell commands and Python snippets that directly interpolate user-supplied file paths and glob patterns without sanitization.\n
Evidence: Phase 1 executes python3 scripts/check-whitespace.py --fix <target-file-or-dir>.\n
Evidence: Phase 3 executes python3 -c "import yaml; yaml.safe_load(open('<file>').read().split('---')[1])".\n
Risk: Maliciously crafted paths (e.g., using shell metacharacters or Python quote escapes) can lead to arbitrary command execution or code injection.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted markdown files and is explicitly instructed to "KEEP" all instructions found within them.\n
Ingestion points: File content read via the Read or Glob tools in Phase 1 and 2.\n
Boundary markers: None specified. The skill does not isolate processed content or instruct the agent to ignore embedded commands.\n
Capability inventory: The skill has Bash, Edit, and Write capabilities, allowing significant filesystem and system impact if subverted.\n
Sanitization: No sanitization or filtering of instructions is performed before they are preserved in the output.\n
Risk: Instructions in target files could manipulate the agent's behavior during or after the condensation process.

condense