The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes and hashes the content of user-controlled shell configuration files (~/.zshrc, ~/.bashrc, etc.) and lists the names of LaunchAgents. The resulting audit report is presented to the agent, which could theoretically contain malicious instructions embedded by an attacker who has already compromised those local files.
Ingestion points: The scripts/env-doctor.sh script reads various shell RC files ($RC_FILES) and the contents of ~/Library/LaunchAgents/ to perform its audit.
Boundary markers: The audit output uses clear section headers, but it lacks specific instructions to the LLM to ignore potentially adversarial content found within the files being audited.
Capability inventory: The skill possesses broad read access to user configuration files and environment state. It has the capability to write to a ledger file (~/.config/env-ledger.md) and a state file (~/.config/env-doctor/state).
Sanitization: The script primarily uses shasum for file integrity checks and ls/awk for listing, which avoids direct execution of the file contents. However, the results (including file names and hashes) are fed back into the agent's context.

env-doctor