The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes logs from external sources (Docker containers), creating a surface for indirect prompt injection if those logs contain malicious instructions designed to influence the agent during analysis.
Ingestion points: Docker logs are retrieved and processed in scripts/fetch-logs.sh.
Boundary markers: The skill does not implement delimiters or explicit 'ignore embedded instructions' warnings when analyzing log data.
Capability inventory: The skill utilizes Read, Write, and Bash capabilities, including commands for database (PostgreSQL), cache (Redis), and container (Docker) management.
Sanitization: No sanitization or escaping is performed on the log content or user-provided incident details before they are interpolated into the final reports in scripts/generate-incident-report.py.
[COMMAND_EXECUTION]: The skill contains instructions and scripts that execute various system-level diagnostic and remediation commands.
Evidence: SKILL.md and referenced scripts like scripts/health-check-all-services.sh use docker exec, docker ps, docker system prune, psql, and redis-cli.
Evidence: There is a discrepancy between the allowed-tools listed in the frontmatter (limited to curl and jq within Bash) and the extensive use of other CLI tools in the instructions, which may lead to execution failures or policy confusion.
Context: While these commands are powerful, they are consistent with the skill's stated purpose of diagnosing and mitigating production service incidents.

incident-response