The Agent Skills Directory

[DATA_EXFILTRATION]: The skill accesses the Goose session database to analyze historical conversation data. This constitutes exposure of sensitive local information as the database contains the full history of interactions with the agent. However, the access is restricted to the local database, no network exfiltration was detected, and the functionality is transparently disclosed as the skill's primary purpose.
Evidence: File scripts/rp_why_baseline.py defines the path to the SQLite database at ~/.local/share/goose/sessions/sessions.db (or the Windows equivalent) and queries conversation messages for analysis.
[INDIRECT_PROMPT_INJECTION]: The skill processes and displays content from historical user prompts. This creates an indirect prompt injection surface where malicious instructions stored in past conversations could be re-processed by the agent during the skill's reporting phase.
Ingestion points: Historical prompt text is read from the messages table in the session database (managed in scripts/rp_why_baseline.py).
Boundary markers: No explicit delimiters or instructions are used to isolate historical content when it is presented in reflection reports.
Capability inventory: The skill can read from the local session database and write to local baseline files in ~/.config/goose/. It has no subprocess or network capabilities.
Sanitization: The skill performs keyword-based classification and truncates displayed prompt text (e.g., to 70 characters for the "Peak Moment" feature in scripts/rp_why_baseline.py).
[EXTERNAL_DOWNLOADS]: The skill is installed from a remote GitHub repository.
Evidence: Documentation in README.md and SKILL.md points to the source repository at https://github.com/block/agent-skills.

rp-why