The Agent Skills Directory

[SAFE]: The skill is entirely comprised of Markdown instructions and documentation templates. It does not include any scripts (Python, Node.js, Shell) or binaries.
[COMMAND_EXECUTION]: The skill utilizes basic filesystem tools (Read, Write) to analyze and update project plans. Execution is scoped to documentation directories (docs/plans/, docs/architecture/, docs/threat-models/**) via frontmatter constraints, which aligns with the stated purpose and limits the potential for unauthorized file access.
[INDIRECT_PROMPT_INJECTION]: The skill acts on internal project files that could theoretically contain malicious instructions. However, the risk is negligible as the skill lacks network capabilities to exfiltrate data and its filesystem write permissions are limited to documentation paths. The skill even includes common security failures like prompt injection as examples for the agent to consider during risk modeling.

pre-mortem