The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted content from bug reports, log files, and source code, creating a potential surface for indirect prompt injection.
Ingestion points: The skill takes input from bug description or report path and is instructed to read files along the trace path using cat -n.
Boundary markers: The skill includes a dedicated 'Safety' section that establishes clear operational boundaries, such as 'Until the user asks for a fix, never edit the user's working tree' and 'Explicit approval first for shared-state writes'.
Capability inventory: The skill utilizes subprocess execution for running tests, CLI commands, headless browsers, and git operations (including git bisect).
Sanitization: The skill mitigates risks through a mandatory 'Feedback loop' and evidence-based 'Investigate' phase, requiring the agent to prove defects through controlled experiments and falsifiable hypotheses rather than blindly following input instructions.

hl-debug