The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it ingests and follows instructions from external files.
Ingestion points: Loads task specifications from workspace/tasks/{task_id}.md in Step 0 using a helper script.
Boundary markers: No boundary markers or 'ignore' instructions are provided when parsing the content of the task specification.
Capability inventory: The skill can read local files via read_file.py and execute arbitrary test code via run_tests.py.
Sanitization: While the task_id input is validated against a specific pattern, the content of the file being read is used to define implementation tasks without further sanitization.
[COMMAND_EXECUTION]: The skill performs command-line operations using local Python scripts to facilitate the TDD workflow.
Evidence: Executes python .claude/skills/bmad-commands/scripts/read_file.py and python .claude/skills/bmad-commands/scripts/run_tests.py across multiple workflow steps.
Context: These executions are used for reading requirements and running the test suite, which are standard operations for a development-focused agent skill.

implement-feature