subagent-driven-development — independently scanned and version-tracked by SaferSkills.
SaferSkills independently audited subagent-driven-development (Agent Skill) and scored it 100/100 (green). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 0 high-severity and 0 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.
Findings & checks · 0 flagged
Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
Execute an implementation plan by giving each task a focused implementation agent, then running two review gates before moving on.
Core rule: the main session coordinates. Agents implement or review bounded work. The main session integrates, verifies, and decides when to proceed.
Use this skill when:
If there is no plan yet, use workbench:writing-plans first. If implementation is starting, use workbench:test-driven-development before writing production code.
For each plan task:
Spec compliance reviewer:
Code quality reviewer:
Do not start the code quality reviewer until spec compliance passes.
Implement this plan task:
[paste one checkbox task]
Ownership:
- You may edit: [files or directories]
- Do not edit: [files or directories]
Context:
- Relevant spec or plan excerpt.
- Relevant prior commits or decisions.
Rules:
- Use workbench:test-driven-development.
- You are not alone in the codebase. Do not revert edits made by others.
- Keep the change minimal.
Return:
- Status: DONE, DONE_WITH_CONCERNS, NEEDS_CONTEXT, or BLOCKED.
- Files changed.
- Verification commands and results.
- Concerns, if any.DONE: proceed to spec compliance review.
DONE_WITH_CONCERNS: read the concerns. Fix in-scope correctness, lint, test, or scope concerns before review.
NEEDS_CONTEXT: provide the missing context and continue with the same task.
BLOCKED: change something before retrying. Provide missing context, split the task, use a more capable model, or stop if the plan is wrong.
Default to one implementation task at a time. This keeps review gates simple and avoids shared-state conflicts.
Use workbench:dispatching-parallel-agents only when tasks are truly independent and have disjoint write scopes. If tasks share files, generated outputs, schemas, lockfiles, or ordering dependencies, run them sequentially.
Read-only review agents may run in parallel when they review different artifacts.
Claude Code: use Agent with a focused prompt for implementation and review agents.
Codex: use spawn_agent for independent sidecar tasks when useful. For urgent blocking work, implement locally and preserve the same task loop. For delegated code edits, assign a disjoint write set and tell the worker not to revert others' changes.
After all tasks pass both review gates:
git status and git diff.workbench:verification-before-completion before claiming the branch is complete or ready.~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.