The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from session transcripts and conversation history to create archives and memory candidates.
Ingestion points: Reads conversation history and transcript files located at ~/.claude/projects/<project>/<uuid>.jsonl to curate past events.
Boundary markers: The instructions do not specify the use of delimiters or 'ignore' instructions when processing content from the session transcripts.
Capability inventory: The skill can push data to an archive log and suggest facts for the /helix-memory tool, which may be persisted long-term.
Sanitization: The instructions emphasize quoting prompts 'verbatim' and only trimming large logs, lacking explicit sanitization of potentially malicious instructions embedded in the session history.
[COMMAND_EXECUTION]: The skill uses shell commands like sed and kill to manage the lifecycle of the archive server by targeting specific Process IDs (PIDs) stored in local metadata files (~/.helix/archive/<project>/<slug>/server.json).
[DATA_EXPOSURE]: The skill accesses sensitive session transcript files (.jsonl) which contain the full history of user interactions. While necessary for the 'archive' functionality, this involves the reading of potentially sensitive user data stored in the local environment.

archive