The Agent Skills Directory

[DATA_EXFILTRATION]: The skill accesses sensitive session logs and history files located in ~/.codex/sessions, ~/.claude/projects, and ~/.kimi/sessions. While this is necessary for its auditing purpose, these files often contain sensitive user data and credentials. The skill includes regex-based redaction to mitigate exposure risk.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it ingests untrusted data from session logs and external tool outputs.
Ingestion points: Reads content from ~/.codex/sessions, ~/.agents/otel-collector/data/processed/stats.json, and project-local SKILL.md files.
Boundary markers: Does not utilize explicit delimiters or isolation markers when presenting log snippets to the agent.
Capability inventory: Capable of writing patches to files in ~/dev/agent-skills and executing validation scripts like quick_validate.py and skill_gate.py.
Sanitization: Implements regex-based redaction for emails, tokens, and common secret patterns (e.g., sk-, Bearer ).
[COMMAND_EXECUTION]: The skill executes several local commands and scripts, including scan_codex_sessions.py, quick_validate.py, skill_gate.py, and CLI tools such as claude auth status, rg, and fd. These operations are performed within the local environment for status checking and validation tasks.

skill-refactor