By insightriot
Disciplined AI development workflow integrating GSD's execution orchestration, Agent Skills' quality enforcement, and patterns from the broader Claude Code plugin ecosystem — with a project-complexity calibration router that right-sizes rigor per project tier. Slash-command prefix: /sig:* (Signal is the project brand; sig is the namespace).
Capture-and-route new work — appends a stamped entry to .planning/FUTURE-IDEAS.md (default) after verbatim capture, sensitive-data scrub, and atomic write. Route explicitly with --question (OPEN-QUESTIONS.md) or --milestone [N] (a milestone holding section). Naked /sig:add (no args) asks one question and files the answer to FUTURE-IDEAS. Not phase-gated.
Phase 0 — classify project into SKETCH / FEATURE / SPIKE / FULL and write .planning/PROFILE.md to drive downstream rigor.
Manual state refresh — diffs git log against STATE.md and refreshes it. Default quick mode shows a one-line readout + diff; --context adds decision + open-question capture so a context-clear that follows can re-orient cleanly.
DISCUSS phase — gather implementation decisions through adaptive questioning before planning. Loads idea-refine and spec-driven-development skills.
Claude Code plugin install-state diagnostician. Detects 5 documented failure modes, can generate a remediation shell script (--fix surgical, --reinstall full canonical), and reports healthy or actionable findings. macOS only first ship.
Executes plan tasks with TDD, atomic commits, and context rot prevention. Spawned by /sig:execute for each task in a wave.
Creates executable phase plans with task breakdown, dependency analysis, and vertical slicing. Core agent of the PLAN phase.
Creates project roadmaps with phase breakdown, requirement mapping, and milestone definition. Used during project initialization.
Researches gray-area decisions and returns structured comparison tables. Used when the team needs to choose between viable alternatives.
Surfaces Claude's implicit assumptions about a phase approach. Forces explicit acknowledgment of what's being taken for granted.
Designs stable, well-documented interfaces. Use when designing new API endpoints, defining module boundaries, creating component prop interfaces, or changing existing public interfaces.
Optimizes agent context setup. Use when starting a new session, when agent output quality degrades, when switching between tasks, or when you need to configure rules files and context for a project.
Builds production-quality UIs. Use when building or modifying user-facing interfaces. Use when creating components, implementing layouts, managing state, or when the output needs to look and feel production-quality rather than AI-generated.
Builds in thin vertical slices. Use when implementing any multi-file change, building a new feature from a task breakdown, refactoring existing code, or any time you're tempted to write more than ~100 lines before testing.
Verifies code against official documentation. Use when implementing framework-specific patterns, building boilerplate or starter code, or when the user asks for documented, verified, or correct implementation.
Modifies files
Hook triggers on file write and edit operations
Uses power tools
Uses Bash, Write, or Edit tools
Own this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimOwn this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimBased on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
Disciplined AI development workflow with a project-complexity calibration layer. Right-sized rigor per project tier — so you don't over-engineer throwaways or under-engineer production systems.
Signal is a Claude Code plugin. Run a six-phase workflow (calibrate → discuss → plan → execute → verify → review → ship) where every phase reads a PROFILE.md written at calibration time, and that drives whether you get TDD, four research agents, an 8-dimension plan validation, a full security audit, all of those — or none of them.
A 30-line throwaway script doesn't need ceremony. A production service handling other people's data does. Signal calibrates once and the workflow self-tunes.
Most AI-development workflows pick a fixed level of rigor. Heavy frameworks demand TDD + plans + reviews on every change; light frameworks skip discipline even when stakes are real. Both fail at the wrong end. Signal asks five diagnostic questions, classifies the work into one of four tiers (SKETCH / FEATURE / SPIKE / FULL), and then every downstream phase reads that classification as its first action.
In Signal's own dogfood:
| Tier | Throwaway | Wall clock | Source LOC | Tests | Research agents | .planning/ artifacts |
|---|---|---|---|---|---|---|
| SKETCH | CSV-to-JSON one-shot | ~5 min | ~30 | 0 (smoke only) | 0 | 8 |
| FULL | Production URL shortener | ~2 hours | ~600 | 39 | 4 | 13 |
Same tooling. Same commands. The difference is calibration.
That FULL run is committed as a worked example you can read end to end: examples/url-shortener/ — the complete .planning/ chain (calibrate → ship) plus a runnable, zero-dependency service. Its README walks through why the project calibrated FULL and what each phase produced.
| Component | Notes |
|---|---|
| Node.js 22+ | ESM, no native modules |
| Claude Code 2.1.141+ | That release shipped the CLAUDE_CODE_PLUGIN_PREFER_HTTPS env var used by the troubleshooting workaround below |
| Operating system | Verified on macOS; Linux/WSL untested — see docs/install-verification.md for the verification matrix |
| Git | Any modern version (.planning/ requires a git repo to be useful) |
/plugin marketplace add insightriot/signal
/plugin install sig@signal
/reload-plugins
Then try /sig:calibrate to confirm the install. Signal's /sig:* commands should autocomplete.
If /plugin install fails with Permission denied (publickey) or Could not read from remote repository, your machine has a non-default SSH config (multi-identity hosts, IdentitiesOnly yes, or no default Host github.com) and Claude Code's plugin loader is reaching for SSH. Two documented escape hatches:
Use Anthropic's HTTPS-prefer env var (Claude Code 2.1.141+):
export CLAUDE_CODE_PLUGIN_PREFER_HTTPS=1
Then retry the install. This is the cleanest fix and doesn't change your SSH config.
Or install from a local clone (no marketplace fetch involved):
GIT_CONFIG_GLOBAL=/dev/null git clone https://github.com/InsightRiot/signal ~/signal
Then in Claude Code: /plugin marketplace add ~/signal → /plugin install sig@signal.
As of v0.1.1, Signal's marketplace.json pins HTTPS explicitly, so option 1 should rarely be needed — but it's there if some other plugin in your stack triggers the same path.
For other install symptoms — /plugin install reporting "already at latest" while running stale code, missing Uninstall verb in the /plugin UI, plugin staying Disabled after reinstall, pre-rename signal@signal cache orphans — see docs/install-troubleshooting.md.
git clone https://github.com/InsightRiot/signal
cd signal
npm install
node tools/validate-plugin.js # should report all checks green
npm test # 777 tests should pass
Then point Claude Code at the local plugin directory (settings → plugins → load from path), or symlink into your ~/.claude/plugins/ if you prefer.
${CLAUDE_PLUGIN_ROOT} for development. Several command markdown files reference paths like ${CLAUDE_PLUGIN_ROOT}/skills/... and ${CLAUDE_PLUGIN_ROOT}/state/config.json. Claude Code sets this env var when the plugin is installed via the marketplace. For local development (or when dogfooding Signal-on-Signal), set it manually before running Claude Code:
export CLAUDE_PLUGIN_ROOT="$(pwd)" # from the Signal repo root
Without it, command markdown still works — Claude resolves the literal path from context — but explicit is faster.
npx claudepluginhub insightriot/signal --plugin sigComprehensive skill pack with 66 specialized skills for full-stack developers: 12 language experts (Python, TypeScript, Go, Rust, C++, Swift, Kotlin, C#, PHP, Java, SQL, JavaScript), 10 backend frameworks, 6 frontend/mobile, plus infrastructure, DevOps, security, and testing. Features progressive disclosure architecture for 50% faster loading.
Harness-native ECC operator layer - 67 agents, 271 skills, 92 legacy command shims, reusable hooks, rules, selective install profiles, and production-ready workflows for Claude Code, Codex, OpenCode, Cursor, and related agent harnesses
Develop, test, build, and deploy Godot 4.x games with Claude Code. Includes GdUnit4 testing, web/desktop exports, CI/CD pipelines, and deployment to Vercel/GitHub Pages/itch.io.
Tools to maintain and improve CLAUDE.md files - audit quality, capture session learnings, and keep project memory current.
Comprehensive PR review agents specializing in comments, tests, error handling, type design, code quality, and code simplification
Access thousands of AI prompts and skills directly in your AI coding assistant. Search prompts, discover skills, save your own, and improve prompts with AI.