The Agent Skills Directory

[COMMAND_EXECUTION]: The skill makes extensive use of the bd CLI (Beads) to manage project state, epics, and task lifecycles (e.g., bd prime, bd mol pour). It also executes local Python scripts via uv run (specifically resolve-feature.py) located in a sibling directory (../dstack-core). These operations are necessary for its documented functionality of translating plans into executable task graphs.
[EXTERNAL_DOWNLOADS]: The instructions reference npx skills update as a mechanism for resolving version staleness. This involves a well-known service (NPM) and is a standard procedure for maintaining development tools.
[PROMPT_INJECTION]: The skill is subject to indirect prompt injection because it ingests untrusted project data (e.g., meeting transcripts, existing documentation files, and source code) and uses this information to drive the agent's planning logic and task creation.
Ingestion points: Reads docs/src/SUMMARY.md, docs/src/planned-features.md, project source code, and user-supplied meeting notes/transcripts.
Boundary markers: None. There are no explicit instructions provided to the agent to treat the content of these files as data only or to ignore embedded instructions.
Capability inventory: The skill has access to Bash, Write, Edit, and Read tools, which could be leveraged if a malicious project file successfully injected instructions.
Sanitization: None. The skill is instructed to process decisions chronologically and copy durable decisions directly into new design.md files.

plan-features