The Agent Skills Directory

[SAFE]: The skill outlines a standard professional debugging methodology with no malicious patterns detected.
[CREDENTIALS_SAFE]: The investigation toolkit explicitly instructs to "Never log secrets, tokens, or personal data" and to "Redact" when comparing environment variables.
[DATA_EXPOSURE]: Instructions prioritize local reproduction and read-only inspection, specifically warning against running destructive or production-mutating commands.
[COMMAND_EXECUTION]: Uses standard development tools like git, node, and bun for diagnostic purposes within the local environment.
[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface (ingesting logs/traces and executing diagnostic commands), but it is mitigated by instructions emphasizing deterministic reproduction and verification.
Ingestion points: User-provided error messages and stack traces (SKILL.md).
Boundary markers: Instructions to quote exact error text and read code paths (SKILL.md).
Capability inventory: File system access, local command execution (git, node, bun), and network diagnostics (curl) (investigation-toolkit.md).
Sanitization: Requirement to remove all instrumentation and redact sensitive data (investigation-toolkit.md).

debug