By heurema
Run a contract-first development pipeline that produces verifiable proof artifacts. Automatically define contracts, implement code changes, audit with three independent AI models (Claude, Codex, Gemini), and package results into a proof pack for CI. Supports Python, JS/TS, and Rust projects with integrated linting and testing.
Bootstrap project context (project.intent.md and project.glossary.json) from an existing codebase using deterministic scan + LLM synthesis + interactive editing. Use --harness to scaffold additional repo-level harness docs.
If the user's task is using Signum init command syntax instead of a feature request - for example:
**Goal:** Bundle all artifacts into a self-contained, verifiable proof package (schema v4.8) with embedded artifact contents.
Display to the user:
If the user's task is exactly `explain` (case-insensitive), do NOT run the pipeline. Instead, output this JSON and stop:
Parses a user feature request into a structured contract.json. Scans codebase for scope signals and risk assessment. Read-only -- never writes code files, only generates contract.json.
Implements code changes according to a contract.json specification. The ONLY agent in Signum that writes code. Includes a repair loop: generate -> check -> fix -> check (max 3 attempts).
Synthesizes project.intent.md and project.glossary.json from deterministic scan signals. Uses ranked source hierarchy and explicit-only Non-Goals extraction. Emits per-section evidence comments and confidence annotations. Read-only — never writes files directly (presents draft for user confirmation).
Semantic code reviewer using Claude Opus. Part of the multi-model audit panel. Analyzes diff against contract for bugs, security issues, and logic errors. Read-only -- never modifies code.
Combines multi-model review results into a consensus verdict. Reads review outputs from Claude, Codex, and Gemini, plus mechanic report. Applies deterministic synthesis rules to produce final audit decision. Read-only -- never modifies code.
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.
Signum is a contract-first proof gate for agentic software changes: it turns a task into a reviewed contract, executes against that contract, audits the result, and packages evidence that humans and CI can inspect.
Signum is an active, experimental baseline for local deterministic development and review workflows. It is not a production certification system.
The canonical runtime docs are:
commands/signum.md for the main CONTRACT → EXECUTE → AUDIT → PACK pipeline.commands/init.md for project bootstrap with /signum:init.This README is an entry point, not the complete runtime specification. Use docs/reference.md when exact behavior matters.
Signum follows one mental model:
CONTRACT → EXECUTE → AUDIT → PACK
First it defines acceptance criteria and scope, then it implements against that contract, audits deterministic and model-assisted evidence, and writes a proofpack for review or CI gating.
Signum expects this local toolchain:
bash
git
jq
python3
For Claude Code, install the plugin from the Emporium marketplace:
claude plugin marketplace add heurema/emporium
claude plugin install signum@emporium
claude "/signum explain"
For Codex App, Signum ships plugin metadata in this repository: .codex-plugin/plugin.json, .agents/plugins/marketplace.json, and platforms/codex/.codex-plugin/plugin.json. In Plugins -> Add marketplace, use:
Source: heurema/signum
Git ref: main
Sparse paths: leave blank
After the marketplace is added, it appears in the Plugins source dropdown as Heurema; the installable plugin inside it is Signum. For fuller setup and platform notes, see QUICKSTART.md.
Run the main pipeline with a task:
/signum "your task"
Signum asks for contract approval before execution. Normal run artifacts are written under:
.signum/contracts/<contractId>/
| Command | Purpose |
|---|---|
/signum "<task>" | Run the canonical Signum pipeline. |
/signum explain | Explain the current pipeline and artifact model. |
/signum archive [contractId] | Archive a completed contract; uses the active contract when no ID is provided. |
/signum close [contractId] | Close or abandon a contract without generating a proofpack; uses the active contract when no ID is provided. |
/signum:init | Bootstrap project context files. |
/signum:init --harness | Bootstrap project context and scaffold repo-level harness docs. Requires Signum >= v4.18.0. |
/signum:init --force | Overwrite existing bootstrap files. |
/signum:init --project-root <path> | Run bootstrap against a specific project root. |
Use the colon form for init commands: /signum:init is the canonical init surface.
.signum/contracts/<contractId>/ is the canonical active contract artifact root. It holds run evidence such as contract.json, combined.patch, audit_summary.json, and proofpack.json.
Root .signum/ is a registry/state/archive namespace with compatibility helpers; normal runs do not create root artifact files or root runtime dirs there.
For compatibility, resume checks use the registry first, with root .signum/contract.json only as a legacy import signal.
For the detailed artifact inventory, see docs/artifact-path-inventory.md and docs/reference.md.
Run the CI wrapper with:
bash lib/signum-ci.sh
The GitHub Actions workflow template lives at:
lib/templates/signum-gate.yml
lib/signum-ci.sh maps proofpack decisions to exit codes:
| Exit code | Decision |
|---|---|
0 | AUTO_OK |
1 | AUTO_BLOCK |
78 | HUMAN_REVIEW |
The template uploads proofpack artifacts and comments the decision on pull requests. See lib/signum-ci.sh, lib/templates/signum-gate.yml, and docs/reference.md for the full CI behavior.
docs/README.md — documentation index separating runtime references, maintainer docs, and historical context.QUICKSTART.md — setup and first-run walkthrough.examples/README.md — small, validator-backed examples for proofpacks, CI gating, and contract shape.docs/how-it-works.md — pipeline narrative and phase details.docs/reference.md — canonical reference for behavior, artifacts, and schemas.docs/api-reference.md — concise integration index for schemas and deterministic scripts.docs/migration-notes.md — compatibility notes for historical artifact roots, proofpack schema versions, and init command naming.docs/RELIABILITY.md — reliability notes and critical journeys.docs/SECURITY.md — trust boundaries and security review triggers.ARCHITECTURE.md — system overview and component map.npx claudepluginhub heurema/signumDaily curated news digest for your domain. One command to set up, zero API keys, works offline after first fetch.
Multi-AI orchestrator — dispatch to Codex CLI and Gemini CLI for review, ask, implement, panel, quorum, continue, diverge, and doctor modes. Diverge generates 3 independent implementations in isolated worktrees with different strategy hints, then compares and selects the best solution. Doctor checks provider health and reports partial success when some providers are unavailable.
Code transformation: Dev SDLC orchestrator (code-shipping pipeline), plan, assert, audit, review, test, refactor, debug, for-sure. Hosts engineering agents.
Plan-first contract engineering for Claude Code and Codex.
Verification-first engineering toolkit for Claude Code. 15 skills across a 5-phase spine (Investigate → Design → Implement → Verify → Ship), 8 specialist agents, an interactive setup wizard. Every skill has rationalizations + evidence requirements. Built for senior ICs and tech leads.
AI-First SDLC — zero-debt development with validators, enforcement, and workflows
Personal Claude Code + Codex dev stack: security hooks, AI-first code conventions, /security-review, /repo-map, /stack-check, portable statusline. Designed to complement other skills-based plugins, not replace them.
Persona-driven AI development team: orchestrator, team agents, review agents, skills, slash commands, and advisory hooks for Claude Code