The Agent Skills Directory

[PROMPT_INJECTION]: The skill presents a surface for indirect prompt injection by processing project-controlled design and task files.\n
Ingestion points: The skill reads external files like design.md and tasks.md to determine implementation tasks and constraints (SKILL.md).\n
Boundary markers: No clear delimiters or instructions are provided to the agent to treat content from these files as untrusted or to ignore instructions embedded within them.\n
Capability inventory: Authorized tools include Bash, Write, Edit, and Read, which grant the skill significant power to modify the environment based on potentially malicious data.\n
Sanitization: Content from ingested files is not sanitized or validated before being used to guide the implementation process.\n- [COMMAND_EXECUTION]: The skill uses the Bash tool to perform complex repository operations and execute local logic.\n
Repository operations: Executes multiple git commands for managing worktrees, checking status, and tracking changes.\n
Local script execution: Uses uv run and python3 to run internal project utilities such as resolve-feature.py and beads-workflow-lock.py for state management.\n
Logic encapsulation: Defines and runs a complex shell function, finalize_feature_interactions, which manages file system writes and audit trail commits to .beads/interactions.jsonl.

implement-feature