The Agent Skills Directory

[SAFE]: The skill implements a robust 6-phase debugging methodology, focusing on establishing feedback loops and verifying fixes. All recommended actions are standard developer workflows.
[COMMAND_EXECUTION]: The instructions involve running local tests, scripts, and CLI tools (e.g., curl, git bisect, npm install). This is the intended purpose of the skill and does not include execution of untrusted or remote code.
[PROMPT_INJECTION]: The skill processes user-provided input through the scripts/hitl-loop.template.sh script, presenting an indirect prompt injection surface.
Ingestion points: User-provided responses are captured via the read command in scripts/hitl-loop.template.sh and printed to stdout for the agent to parse.
Boundary markers: The script uses a --- Captured --- header as a delimiter for the agent to identify the results section.
Capability inventory: Across its files, the skill expects the agent to execute shell commands, read/write local files, and perform network requests to local dev servers.
Sanitization: The bash script uses printf -v to safely assign inputs to shell variables, preventing direct command injection into the helper script, though the content itself is not sanitized before being returned to the agent context.

diagnosing-bugs