The Agent Skills Directory

[COMMAND_EXECUTION]: The pack-export.sh script orchestrates complex workflows by executing shell commands and invoking sibling scripts (funnel.py, curate-sessions.sh, pick-sessions.sh) using bash and python3.
[DATA_EXFILTRATION]: The skill is designed to scan and export user session history. It implements a robust regex-based scanner (CRED_RE) to identify sensitive credentials (e.g., AWS AKIA, GitHub tokens, Slack keys, and RSA Private Keys) to prevent accidental data leakage. By default, sessions containing these patterns are blocked from export unless the user explicitly uses the --allow-credentials flag.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits the following vulnerability surface for indirect prompt injection:
- Ingestion points: Processes local session history files (byte-identical copies of user history) in the scripts/pack-export.sh driver.
- Boundary markers: Boundary markers are not used as the script filters data for archival rather than interpolating it into active LLM prompts.
- Capability inventory: Includes file system read/write access, subprocess execution (Python and Bash), and archive generation for data delivery.
- Sanitization: Utilizes a credential detection regex to flag and exclude sensitive content from exported archives.

trajectory-packs