The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect instructions by processing external data logs.
Ingestion points: The agent ingests status.message, metrics.jsonl, and trainer logs to identify job failures as described in references/triage-paths.md and references/tito-failures.md.
Boundary markers: The skill does not define specific delimiters or warnings to separate log content from agent instructions during processing.
Capability inventory: The skill utilizes the firectl CLI, pip for package management, and python3 for session management as shown in SKILL.md and references/triage-paths.md.
Sanitization: The skill contains instructions to redact credentials and customer identifiers from evidence bundles before escalation in SKILL.md.
[COMMAND_EXECUTION]: The skill uses local CLI tools and package managers for diagnostic data collection.
Evidence: The skill uses firectl to retrieve job metadata and pip to ensure the presence of the fireworks-ai library in references/triage-paths.md.
[DYNAMIC_EXECUTION]: The skill executes short dynamic scripts to manage session state.
Evidence: The skill employs a python3 one-liner to generate a unique session UUID for telemetry purposes in SKILL.md and references/triage-paths.md.

debug