The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill possesses a significant attack surface for indirect prompt injection because it is designed to ingest and process untrusted external data which then influences high-privilege actions like file modification and command execution.
Ingestion points: The skill reads user-provided bug descriptions ($ARGUMENTS), source code files during path tracing (SKILL.md, Phase 1), and log files during investigation (reference/investigation-ladder.md, Rungs 2 and 3).
Boundary markers: Absent. The instructions do not specify the use of delimiters or protective instructions to prevent the agent from following commands embedded within the code or logs being analyzed.
Capability inventory: The skill allows the agent to read files, write modifications to the filesystem (SKILL.md, Phase 5), and execute shell commands for testing and debugging (SKILL.md, Phase 5; reference/investigation-ladder.md, Rung 5).
Sanitization: Absent. There is no evidence of content sanitization or validation before the processed data is used to formulate hypotheses or fixes.
[COMMAND_EXECUTION]: The skill requires the agent to execute system commands to facilitate the debugging process.
Evidence: The workflow instructs the agent to check git status (SKILL.md, Phase 1).
Evidence: The investigation ladder specifies using git bisect to identify the source of regressions (reference/investigation-ladder.md, Rung 5).
Evidence: The skill mandates that the agent "run tests" after applying any code fixes (SKILL.md, Phase 5).

debug