The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from PostHog tool outputs and processes it through local Python helper scripts. If event names or properties in the PostHog platform contain malicious instructions, they could influence the agent's behavior.
Ingestion points: Data returned by posthog:query-trends, posthog:query-funnel, posthog:execute-sql, and the resulting analysis from scripts/compare_to_prior_periods.py and scripts/breakdown_attribution.py.
Boundary markers: Absent. The agent is instructed to treat the output of these tools and scripts as factual context for its report.
Capability inventory: The agent can execute custom SQL, create annotations, and create new insights within the PostHog platform.
Sanitization: None detected. The helper scripts convert JSON data directly into Markdown strings without filtering for potential injection patterns.
[COMMAND_EXECUTION]: The skill requires the agent to execute shell commands to perform data analysis and inspect the local environment.
Evidence: SKILL.md explicitly instructs the agent to run Python scripts (python3 scripts/compare_to_prior_periods.py) and utilizes git log to check for code changes that might align with metric anomalies.
[DATA_EXPOSURE]: The use of the git log command allows the agent to access sensitive repository metadata.
Evidence: The skill uses git log to correlate code changes with data anomalies, which grants the agent access to commit messages, author names, and timestamps from the host environment's git repository.

investigate-metric