The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from external local files, which could contain untrusted content that influences the agent's behavior. \n- Ingestion points: The skill reads shinkoku.config.yaml, .shinkoku/progress/progress-summary.md, and .shinkoku/progress/02-assess.md to initialize its workflow (SKILL.md). \n- Boundary markers: There are no explicit instructions or markers to distinguish between trusted instructions and potentially untrusted data within these files. \n- Capability inventory: The agent has file write access via the 'Write' tool and is instructed to interact with local CLI scripts import_data.py and ledger.py (SKILL.md). \n- Sanitization: The skill does not mention validation or sanitization steps for the data retrieved from the configuration or progress files. \n- [COMMAND_EXECUTION]: The instructions direct the agent to execute specific Python scripts with variable parameters. \n- Evidence: SKILL.md includes logic for running import_data.py payment-statement --file-path PATH and ledger.py add-business-withholding --db-path DB --input FILE. While these appear to be internal tools belonging to the shinkoku suite, they constitute a command execution surface managed by the AI agent.

gather