The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill is designed to read agent session logs from local paths such as ~/.claude/history.jsonl and ~/.codex/sessions/. These directories contain historical interaction data which may include sensitive source code, internal environment variables, or private information.- [PROMPT_INJECTION]: The skill processes untrusted data from external sources like pull request descriptions, commit history, and session logs, which creates an indirect prompt injection surface.
Ingestion points: Reads files and artifacts specified in the skill arguments, including PR bodies and log files.
Boundary markers: The instructions do not provide delimiters or specific warnings to ensure the agent ignores embedded instructions within the ingested content.
Capability inventory: The skill utilizes shell utilities like jq and grep and generates a local HTML report.
Sanitization: The skill lacks instructions for validating or escaping external content before it is processed by the agent.- [COMMAND_EXECUTION]: The skill employs shell commands such as jq, grep, sort, and uniq to analyze log files. These commands are executed on data that may originate from untrusted sources.- [EXTERNAL_DOWNLOADS]: The skill fetches extension guidelines from the agent's official documentation, such as Claude Code's documentation at code.claude.com/docs.

opportunity-scan