The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted project documentation such as specifications, plans, and configuration files to drive code generation and command execution.
Ingestion points: Reads markdown files from 02-DOCS/wiki/sdd/specs/ and 02-DOCS/wiki/sdd/plans/, as well as user profile data from 02-DOCS/wiki/harness/user-profile.md.
Boundary markers: While the skill emphasizes a 'done-check' contract and a project 'constitution', it does not explicitly specify technical delimiters (like XML tags) to isolate ingested content within the prompt.
Capability inventory: The skill has the ability to write to the file system (progress logs, decisions), execute git commands, run vendor tools via npx, and dispatch subagents.
Sanitization: The risk is mitigated by the mandatory Red-Green-Refactor loop, per-task isolated reviews, and human-in-the-loop checkpoints required before proceeding to the next task.
[DYNAMIC_EXECUTION]: The skill generates implementation code and unit tests at runtime and subsequently executes them.
Evidence: The core loop in SKILL.md requires the agent to write a failing test, then write the code to pass it, and execute the test suite (e.g., pytest, go test, vitest) to verify behavior.
Context: This dynamic execution is the primary intended function of the skill and is constrained by the requirement for incremental, test-first development.
[COMMAND_EXECUTION]: The skill invokes local shell scripts and a vendor-specific tool to manage the development environment.
Evidence: The instructions direct the agent to run scripts/task-brief for requirement extraction, scripts/review-package for preparing diffs, and npx @ericrisco/rsc registry refresh for skill management.
Context: The shell scripts include input validation (e.g., verifying git revisions and integer task IDs) to prevent command injection, and the npx package belongs to the skill's author.

implement