Skip to content
Back to skills

Implement

ASecurity

Load a Linear Issue and execute the full workflow from branch creation to implementation, testing, commit, and PR creation

  • 16 stars
  • 0 votes
  • 0 copies
  • 0 views
  • Added September 22, 2026
securitygotestinggit

Works with

  • claude code
  • mcp

Security analysis

A100/100

Scanned September 22, 2026

npx -y skills add usk6666/yorishiro-proxy --skill implement --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Implement?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Implement
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/usk6666-implement/badge)](https://www.skillsdirectory.com/skills/usk6666-implement)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
description: "Load a Linear Issue and execute the full workflow from branch creation to implementation, testing, commit, and PR creation"
user-invokable: true
---

# /implement

An end-to-end workflow skill for implementing a Linear Issue through to PR creation.

## Arguments

- `/implement <Issue ID>` — Implement the specified Issue (e.g., `/implement USK-12`)

## Steps

1. **Load Issue**: Fetch Issue details with `mcp__linear-server__get_issue`
2. **Update Issue status**: Update status to "In Progress"
3. **Create branch**: Generate an appropriate branch name from the Issue
   - feat: `feat/<id>-<short-desc>`
   - fix: `fix/<id>-<short-desc>`
   - other: `chore/<id>-<short-desc>`
4. **Design review** (mandatory for any Issue that introduces a new public type, modifies an interface, or touches more than one package; **skip for** docstring-only / test-only / single-call-site fixes — note the skip reason in the user-facing plan): launch the design-reviewer agent. See "Design Review Invocation" below.
5. **Present plan to user**: Issue summary, file actions (new / modified / untouched), key resolved decisions from the design review, deferred items, unresolved questions. Get user confirmation before implementing. For trivial Issues that skipped Step 4, present a one-paragraph plan instead.
6. **Implement**: Implement code based on the plan
7. **Write tests**: Write tests for the implementation
8. **Verify**:
   - Auto-format with `gofmt -w .`
   - `make lint` (gofmt check + golangci-lint: govet / staticcheck / unused / ineffassign / gocyclo)
   - `make build`
   - `make test`
9. **Commit**: Commit in Conventional Commits format
   - Include `Refs: <Issue ID>` in the commit message footer
10. **Push**: Push to remote with `git push -u origin <branch-name>`
11. **Create PR**:
    - Review changes with `git diff main...HEAD`
    - PR title in Conventional Commits format (e.g., `feat(protocol): add HTTP handler`)
    - PR body follows the template below
    - Create PR with `gh pr create`
12. **Update Issue status**: Update status to "In Review"
13. **Report results**: Display implementation summary + PR URL

## Design Review Invocation

Read `.claude/agents/design-reviewer.md` and replace the placeholders before launching:

| Placeholder | Value |
|---|---|
| `{{SCOPE_DESCRIPTION}}` | Issue title + description + tentative file action list (new / modified) inferred from the Issue |
| `{{SPEC_REFERENCES}}` | Paths to relevant spec/design docs (`docs/rfc/envelope.md` for data path; protocol-specific RFCs; `docs/rfc/plugin-migration.md` for plugin work). Cite specific sections, not whole files. |
| `{{PACKAGES_TO_SURVEY}}` | Packages the new code creates, modifies, or depends on |
| `{{COMPLETED_CONTEXT}}` | One-paragraph summary of relevant existing public surface this Issue will integrate with (types/functions, not the whole architecture) |
| `{{PRODUCT_IDENTITY}}` | Read from `.claude/skills/review-gate/SKILL.md` Phase 1-4 "Product context" block — single source of truth |
| `{{PRINCIPLES}}` | The MITM Implementation Principles from `CLAUDE.md` (the 6-item list). Quote verbatim. |

Launch:
```
Agent(
  description="Design review: <Issue ID>",
  subagent_type="general-purpose",
  prompt=<composed prompt>
)
```

Process the result:
- **All resolved** → fold the resolved decisions table into the Step 5 plan
- **Unresolved items exist** → present **only** the unresolved questions to the user in Step 5 with the agent's proposed answers and trade-offs. Do not ask the user about resolved questions.
- **Fitness check FAIL** → stop and report. Do not proceed to Step 5 until the scope is adjusted.

## PR Body Template

```markdown
## Summary
- <bulleted list of changes>

## Test plan
- [ ] Test items

Resolves <Issue ID>
Linear: https://linear.app/usk6666/issue/<Issue ID>

🤖 Generated with [Claude Code](https://claude.com/claude-code)
```

## Notes

- If build or tests fail, fix the issues before re-running
- Determine implementation scope from Issue labels and description
- The design review in Step 4 already serves as the implementation plan for non-trivial Issues — Step 5 just relays it to the user
- Since build/test verification was done in Step 8, do not re-run when creating the PR
- If PR creation fails, guide the user to manually run `/pr`
- Before completing implementation, verify that the feature is accessible from the config file (no missing config struct, validation, or init function changes)

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.