The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process data from external, potentially untrusted sources like git commits and pull requests to build a project timeline.
Ingestion points: The skill gathers data from audit-trail.skill, version control history (commits, PRs), incident reports, and telemetry logs.
Boundary markers: The instructions do not provide explicit delimiters or "ignore embedded instructions" directives to prevent the agent from obeying commands hidden within the ingested project metadata.
Capability inventory: The skill has the capability to write distilled lessons into the agents/memory/CONTINUITY.md file, which persists across agent sessions and influences future behavior.
Sanitization: While the skill includes instructions to redact PII and credentials, it does not specify any sanitization or filtering to detect or remove adversarial instructions embedded in the source data.

retrospective