The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the gbrain jobs submit shell command to execute ingestion logic and maintenance tasks. The instructions guide the agent to wrap pipeline logic into shell commands for execution by background workers. Evidence is found in SKILL.md (Phase 9: BULK) and MANIFEST-PATTERN.md (Periodic commit during long runs).
[PROMPT_INJECTION]: The skill is designed to ingest large external datasets, which introduces a risk of indirect prompt injection where malicious instructions embedded in the data could influence agent behavior.
Ingestion points: SKILL.md (Description, Phase 2: ACCESS) describes ingesting email takeouts, chat exports, and API dumps.
Boundary markers: The skill defines an untrusted-quoted fence and recommends flagging untrusted_directives: true to separate data from instructions.
Capability inventory: The agent can write to multiple directories (projects/, sources/) and submit background shell jobs.
Sanitization: Instructions recommend detecting agent-directed imperatives at transform time.
[DATA_EXFILTRATION]: The skill automates the transfer of repository state, including ingested artifacts and metadata, to external git remotes via git push. Evidence: MANIFEST-PATTERN.md provides a shell command template for recurring sync operations involving potentially sensitive data.

bulk-ingestion