The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides bash script templates that interpolate the <slug> argument directly into shell commands (e.g., spec_dir=docs/specs/<slug>). If the agent is provided with a maliciously crafted slug containing shell metacharacters (e.g., ; curl ...), it could lead to arbitrary command execution in the user's environment.
[INDIRECT_PROMPT_INJECTION]: The skill's logic relies on reading and parsing project-controlled files to determine which files to delete or move, creating a surface for indirect attacks if those files are modified by an untrusted source.
Ingestion points: The script reads configuration and state from docs/specs/$slug/references/_index.md and multiple task_NN.md files.
Boundary markers: There are no explicit delimiters or instructions provided to the agent to treat the contents of these files as untrusted data.
Capability inventory: The skill has the capability to delete directories (rm -rf), change file permissions (chmod), and create temporary files (mktemp).
Sanitization: The provided awk script performs validation on the _index.md file, checking for correct field counts, specific data types, and preventing path traversal by rejecting .., absolute paths, or path separators in the file path fields.

archive-spec