The Agent Skills Directory

[COMMAND_EXECUTION]: The utility script scripts/deliver-task.mjs executes Git commands using execFileSync to manage worktrees, verify commits, and handle repository snapshots. The script performs validation on inputs such as Task IDs and Git OIDs using regular expressions to prevent command injection.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data, including repository source code and user-provided task contracts (task.json). These inputs serve as the primary source of instructions for the agent's work. The skill implements boundary markers by isolating the environment in Git worktrees and using explicit allowed/forbidden path lists to mitigate risks associated with processing untrusted data.
Ingestion points: The skill reads task definitions from task.json (via stdin), architecture guidelines from ARCHITECTURE.md, and repository source files as described in SKILL.md and scripts/deliver-task.mjs.
Boundary markers: Instructions mandate the use of isolated Git worktrees, a dedicated .dev-task/ state directory, and explicit allowedPaths and forbiddenPaths filtering.
Capability inventory: The agent is authorized to perform file system read/write operations and execute git subprocesses through the provided helper script.
Sanitization: The deliver-task.mjs script performs JSON schema validation and regex-based input filtering for task metadata and Git OIDs.

deliver-task