The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes find-polluter.sh, a shell script that uses the find command to locate test files and executes them using npm test. It also suggests diagnostic instrumentation examples such as security list-keychains and codesign to debug environment issues.
[INDIRECT_PROMPT_INJECTION]: The skill requires the agent to ingest and analyze external data such as error messages, stack traces, and component logs. This creates an attack surface where maliciously crafted logs from a compromised system could influence agent behavior.
Ingestion points: Phase 1 of SKILL.md (error messages, stack traces) and Phase 1, Step 4 (logging data entering/exiting component boundaries).
Boundary markers: None explicitly defined for isolating untrusted log data.
Capability inventory: Execution of shell scripts via find-polluter.sh and interaction with build tools like npm and codesign in SKILL.md examples.
Sanitization: No specific sanitization or filtering of error/log content is mandated before analysis.
[DATA_EXFILTRATION]: While the skill contains examples for inspecting environment variables (env | grep IDENTITY) and macOS keychain identities (security find-identity), these are presented as diagnostic steps for debugging build/signing failures rather than automated exfiltration patterns. No external network transmission of this data was detected.

systematic-debugging