The Agent Skills Directory

[COMMAND_EXECUTION]: The skill requires the agent to generate and execute reproduction scripts or tests to confirm bug behavior. This is a core part of the 'repro-first' debugging workflow.
[INDIRECT_PROMPT_INJECTION]: The skill defines a surface for processing untrusted data.
Ingestion points: Step 1 (Triage) involves gathering logs, screenshots, and context provided by users.
Boundary markers: The instructions do not define specific delimiters for isolating user-provided descriptions or log content.
Capability inventory: The agent can execute scripts/tests (Step 2), perform file-writes via safe-change (Step 4), and update documentation files (Step 5).
Sanitization: No explicit sanitization or filtering of the user-provided symptoms or logs is mentioned.
[DATA_EXPOSURE]: The skill instructs the agent to write incident documentation to local paths such as llmwiki/wiki/sources/ and update tracking files like llmwiki/html/problem-tree.html. These operations are consistent with standard software maintenance and documentation practices.

orca-issue