The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from PostHog error logs (such as error messages and stack traces) which could contain malicious instructions. It mitigates this risk by explicitly instructing the agent to treat all error content as data to be summarized rather than instructions to be followed.
Ingestion points: External error data retrieved via the PostHog MCP server tools (list-errors, error-details) in SKILL.md.
Boundary markers: The skill provides explicit instructions to quote error text verbatim and disregard embedded instructions intended to override agent behavior.
Capability inventory: The skill uses curl, jq, node, and a ./notify utility for data processing and alert delivery.
Sanitization: Instructions include a directive to redact PII (emails, IDs, tokens) and secrets before including them in notifications, referencing PostHog links instead.
[COMMAND_EXECUTION]: The skill utilizes shell commands including bash, cat, jq, and node to format data and generate reports. These operations are necessary for the skill's operation in the described read-only environment where standard file-writing tools are restricted.
[EXTERNAL_DOWNLOADS]: The skill connects to the official PostHog MCP server (mcp.posthog.com/mcp) to fetch error metrics. This is a well-known service, and the connection is used strictly for the skill's primary monitoring function.

posthog-errors