The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data such as error messages, log files, and traces. * Ingestion points: User-provided input via the hitl-loop.template.sh script and reading of captured artifacts (HAR files, core dumps) as described in SKILL.md. * Boundary markers: The skill instructions emphasize manual redaction and the use of environment variables for credentials but do not implement programmatic delimiters for ingested content. * Capability inventory: The agent is authorized to execute commands, write local scripts, and perform network requests (curl) for reproduction. * Sanitization: The skill relies on the agent following explicit redaction instructions rather than automated filtering.
[DYNAMIC_EXECUTION]: The skill methodology involves generating and running transient scripts to isolate and test bugs. * Evidence: SKILL.md encourages the creation of fuzzing loops, differential testing scripts, and throwaway harnesses. These are generated from scratch or based on the provided hitl-loop.template.sh to execute logic on the local system.

diagnosing-bugs