The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and act upon untrusted task descriptions provided at runtime, creating a surface for instructions embedded in task data to influence the agent's behavior.
Ingestion points: External task data enters the agent context through the {task_description} variable in SKILL.md and via individual task files (tasks/task-*.md) processed in run-implementation.md.
Boundary markers: The workflow implements a "Grounding check before diff" and a "Constitution over card" rule, which explicitly instructs the agent to favor repository-wide architectural rules over potentially conflicting instructions in a task card.
Capability inventory: The agent has the capability to read and write files within the repository, perform git operations (git status, git diff), and execute shell commands for unit and integration testing via Docker.
Sanitization: All generated code is subjected to a mandatory "Review loop" involving sibling skills (ns-reviewer and ns-judge) that provide independent validation gates (requiring a score of 10 for approval) before changes are finalized.

ns-coder