The Agent Skills Directory

[DYNAMIC_EXECUTION]: The file scripts/delivery_bridge.py uses importlib.util to load Python modules from paths computed at runtime based on the script's directory. This mechanism integrates the orchestrator with sub-skills like finalize and implement. While it includes path verification to ensure modules are within the expected package hierarchy, dynamic loading from computed paths is a significant execution vector.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from GitHub issue bodies and comments.
Ingestion points: Data enters the context during the intake stage in scripts/workflow.py via issue snapshots.
Boundary markers: The SKILL.md and references/intake.md documentation explicitly label issue content as untrusted data that must not choice executable paths or choose tools.
Capability inventory: The skill conducts file operations for state storage and dynamic module loading; tests include subprocess calls for repository management.
Sanitization: The skill uses structured JSON validation for snapshots and relies on independent host-side identity measurement to verify the state of the workspace.
[COMMAND_EXECUTION]: Test scripts scripts/test_delivery_bridge.py and scripts/test_workflow.py utilize subprocess.check_output to execute git commands. Although these operations are limited to test files, they represent the direct execution of shell commands within the codebase.

resolve-issue