The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and act upon data from prior agent sessions, which serves as a potential vector for malicious instructions embedded in those records.
Ingestion points: The skill reads prior trails from local workspace transcripts, links to other agent runs, and pause notes stored in .scratchpad/<slug>/resume.md.
Boundary markers: There are no specific instructions or delimiters provided to distinguish between legitimate session data and potentially malicious instructions within the ingested content. The skill is explicitly told to "Treat the prior trail as authoritative input."
Capability inventory: The skill possesses the ability to execute git commands (git log, git diff, git commit), write to the file system (.scratchpad/), and invoke other agent skills such as autonomous-run or diagnosing-bugs.
Sanitization: The instructions do not mandate any sanitization or filtering of the content retrieved from transcripts or external links before it is processed as authoritative context.

pause-resume