The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses a Python script (check_layout.py) that invokes a headless browser (Chrome/Chromium/Edge) via subprocess.run to validate the layout of generated HTML reports. This is a local operation intended for quality assurance and uses standard subprocess call patterns without shell interpolation.
[INDIRECT_PROMPT_INJECTION]: The skill ingests raw conversation history from local session logs. Since this data contains untrusted dialogue from previous AI interactions, there is a theoretical risk that instructions embedded in past sessions could influence the analyzing agent.
Ingestion points: Session logs are collected by scripts/collect.py from ~/.claude/projects and ~/.codex/sessions, then read by the AI from messages.json and messages.txt.
Boundary markers: No specific boundary markers or instruction-isolation delimiters are defined for the message content during the analysis phase.
Capability inventory: The agent can generate new SKILL.md files and execute local scripts.
Sanitization: Messages are truncated for the text summary, but the agent accesses full raw text from JSON files without prompt-injection sanitization.

session-review