The Agent Skills Directory

[SAFE]: The skill includes three Python helper scripts (scripts/assess_health.py, scripts/fleet_survey.py, and scripts/render_run_report.py) that are used to format JSON data fetched from the PostHog platform. A detailed audit confirms these scripts are pure formatters; they rely exclusively on the Python standard library, perform no network I/O, and limit file operations to reading user-provided data files.
[INDIRECT_PROMPT_INJECTION]: The skill processes summaries and logs from scout runs, which contain AI-generated reasoning and tool outputs. This represents an ingestion surface for untrusted data that could potentially contain instructions aimed at influencing the agent's behavior. However, because the skill is strictly read-only and exploratory, and the data is processed through local formatting scripts, the potential for exploitation is minimal and inherent to the scout observability use case.
Ingestion points: Data enters the agent context via scout-runs-list, scout-runs-retrieve, and tasks-runs-session-logs-retrieve (e.g., in SKILL.md).
Boundary markers: No specific boundary markers or 'ignore embedded instructions' directives are mandated when the agent interprets prose summaries.
Capability inventory: All tools and scripts are read-only; no capabilities for writing, modifying, or exfiltrating project data are present in this skill.
Sanitization: The agent is instructed to use the provided Python scripts for reconstruction and formatting, which provides a structured way to handle the data, though no explicit sanitization of the summary text is performed.

exploring-scouts