The Agent Skills Directory

[PROMPT_INJECTION]: The skill is designed to process external, untrusted data (CI logs and error reports), which introduces a surface for indirect prompt injection attacks where malicious instructions could be embedded in the logs to influence the agent's behavior.
Ingestion points: External CI logs and error reports processed by the classification and ticket generation steps (SKILL.md).
Boundary markers: Absent; the provided LLM prompts do not use specific delimiters or instructions to ignore embedded commands in the input data.
Capability inventory: The skill documents the use of the GitHub CLI (gh) to create and manage issues, providing a potential action vector for successful injections (SKILL.md).
Sanitization: While the skill performs normalization to strip timestamps and IDs for stability, it does not sanitize the logs for potential instruction-based injection.
Mitigation: The skill explicitly mandates human review and approval (Step 7) before any automated actions are taken, significantly reducing the risk of exploitation.
[COMMAND_EXECUTION]: The skill provides integration examples using standard command-line tools like the GitHub CLI (gh) and jq. These commands are used to automate ticket creation and are consistent with the skill's stated purpose of defect management.

ai-bug-triage