The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill operates as an orchestrator, processing outputs from multiple subagents (Reviewer, Builder) and external test suites. This multi-step pipeline ingests potentially untrusted data from repository code and subagent messages, creating a surface for indirect prompt injection attacks.
Ingestion points: The skill ingests data from Reviewer and Builder agent outputs, council rulings, and repository files during the 'land' and 'ship' pipelines (SKILL.md, pipelines.md).
Boundary markers: The skill employs structured 'briefs' and standardized return formats to delimit communications between agents, which helps mitigate accidental instruction following (briefs.md).
Capability inventory: The skill is capable of spawning subagents, modifying local files, executing arbitrary test suite commands, and performing remote Git operations like pushing commits and managing Pull Requests (SKILL.md, pipelines.md).
Sanitization: The instructions include critical safety checks, requiring the agent to manually verify reviewer claims against the source code and mandates human approval for deployment or other irreversible actions (SKILL.md, pipelines.md).
[COMMAND_EXECUTION]: The skill facilitates the execution of test suites and static analysis tools ('gates') as well as Git version control commands. These are primary functions of the skill and are constrained by authority rules requiring user permission for remote effects.
[EXTERNAL_DOWNLOADS]: The 'ship' pipeline involves dependency installation and fixture setup, which is a standard part of the development workflow but involves fetching external code from package registries.

do