The Agent Skills Directory

[SAFE]: The skill's logic and tools are consistent with its stated goal of performing project 'premortems'. No malicious behaviors or safety bypasses were found.
[COMMAND_EXECUTION]: The skill instructs the agent to use the Bash tool for benign reporting tasks, such as retrieving the current date for filenames and opening the generated HTML report locally.
[DATA_EXFILTRATION]: There are no network requests or attempts to access sensitive system files like SSH keys or environment secrets. It only accesses files related to the project context (e.g., CLAUDE.md, project briefs).
[PROMPT_INJECTION]: The skill has an indirect prompt injection surface.
Ingestion points: Reads project files (e.g., CLAUDE.md, project briefs) in Step 1.
Boundary markers: None are explicitly used when interpolating plan details into sub-agent prompts in Step 4.
Capability inventory: Uses Bash, Write, Read, Glob, Grep, and Agent tools across various steps.
Sanitization: No sanitization or escaping of the ingested project data is performed before prompt interpolation. This surface is considered a low-risk functional requirement for the skill's intended use.

premortem