The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The init workflow instructions require the agent to fetch installation documentation from https://github.com/gastownhall/beads at runtime. This involves accessing an external source that is not listed as a trusted vendor.
[REMOTE_CODE_EXECUTION]: By instructing the agent to fetch the install docs at run time rather than guessing and follow the documented installation methods (such as running an install script), the skill facilitates the execution of remote code from an untrusted third-party repository.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection by ingesting untrusted data from mission.md and prior task outputs.
Ingestion points: mission.md and metadata.research_step.output are read across multiple workflows.
Boundary markers: No delimiters or explicit instructions to ignore embedded commands are present when processing these inputs.
Capability inventory: The skill can create and update issues with arbitrary metadata using the bd CLI and execute bash scripts.
Sanitization: No sanitization or validation of the content of mission.md or task outputs is performed before they are used to drive subsequent planning or execution steps.
[COMMAND_EXECUTION]: The skill performs environment setup and state management through several shell scripts and system utilities, including package managers like Homebrew and apt-get.

asta-flows