The Agent Skills Directory

[SAFE]: The skill implements a comprehensive and disciplined debugging methodology. The core instructions (Iron Law) require the agent to find root causes and verify all fixes with fresh evidence, which serves as a significant safety guardrail against hallucinated or incorrect fixes.
[COMMAND_EXECUTION]: The skill includes a bash script scripts/find-polluter.sh that automates bisection to identify tests creating unwanted state (e.g., leftover directories). It uses standard shell commands like find and npm test on local files. This behavior is appropriately scoped to local development and debugging purposes.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external inputs such as error logs and stack traces. While this presents an ingestion surface for untrusted data, the skill's verification protocols (described in references/verification.md) require independent confirmation of all claims, effectively mitigating risks from potentially malicious instructions embedded in error messages.
Ingestion points: Error logs, stack traces, and test outputs (referenced in references/systematic-debugging.md).
Boundary markers: The references/verification.md file defines an "Iron Law" requiring fresh evidence before any claim is made, acting as a logical boundary against following instructions in input data.
Capability inventory: Execution of local tests via npm test in scripts/find-polluter.sh.
Sanitization: The process relies on verification of the final state rather than trust in the input content.

Debugging