The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data including error messages, stack traces, application logs, and external API responses.
Ingestion points: The agent is instructed to read stack traces, log files, and API response bodies in SKILL.md (Stage 1) and references/playbooks.md.
Boundary markers: The skill employs a rigid four-stage framework with explicit task transitions, which acts as a logical boundary.
Capability inventory: The skill utilizes file system reading, git operations, and network testing tools (curl, httpie).
Sanitization: No specific sanitization of input data is mentioned, as the focus is on manual and guided analysis rather than automated processing.
[COMMAND_EXECUTION]: The skill instructs the agent to use various shell commands for diagnostic purposes.
Evidence: SKILL.md and references/evidence-patterns.md suggest using git diff, git log, git bisect, and shell loops (for i in {1..100}; do ...) for reproduction and analysis. These are standard developer tools used appropriately for the skill's primary purpose.

debugging