The Agent Skills Directory

[SAFE]: The skill instructions in SKILL.md and references/debug-policy.md establish a highly restrictive and secure workflow. it explicitly forbids automatic code patching, requires human approval for all changes, and focuses on minimal diagnosis over broad refactoring.- [SAFE]: The implementation in scripts/safe_debug.py is secure and uses only standard python libraries. it performs local analysis of error symptoms through string matching and writes findings to a local directory without any network access, subprocess execution, or dynamic code evaluation.- [SAFE]: Potential indirect prompt injection surfaces are handled responsibly. 1. Ingestion points: scripts/safe_debug.py reads external error content via --error-file or --error-text. 2. Boundary markers: the script encapsulates error excerpts within markdown code blocks in the generated diagnosis reports. 3. Capability inventory: the skill has no network, shell execution, or high-privilege capabilities. 4. Sanitization: input is treated as raw text for classification purposes with no interpolation into dangerous contexts.

safe-debug