The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The setup workflow in workflows/init.md directs the agent to install the bd (beads) command-line tool and jq. It instructs the agent to consult an external repository (https://github.com/gastownhall/beads) at runtime to find and follow installation instructions, which can involve executing remote scripts or using package managers from an untrusted third-party source.
[EXTERNAL_DOWNLOADS]: The skill setup requires downloading software and documentation from the repository github.com/gastownhall/beads, which is not a recognized trusted vendor or well-known service.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface (Category 8).
Ingestion points: The workflow processes user-controlled content from mission.md and structured task outputs stored in the bd tool.
Boundary markers: The skill uses YAML schemas (assets/schemas.yaml) and structural validation scripts (scripts/validate-output.sh) to maintain output integrity.
Capability inventory: The agent has permissions to run bash scripts and the bd CLI tool, which allows for file system operations and command execution based on the research state.
Sanitization: Content validation is restricted to structural and type checks; no NLP filtering is performed to identify or neutralize instructions embedded in input data.
[COMMAND_EXECUTION]: The skill extensively uses shell execution to manage its lifecycle, including running custom bash scripts (scripts/*.sh) and the third-party binary bd to update task metadata and summary files.

research-step