ActiveLoom / reusable workflows for coding agents

Give your agents discipline.
Evidence for every PR.

Coding agents write code fast, but without guardrails they regress, hallucinate passing reviews, and burn tokens in loops. ActiveLoom supplies battle-tested skills, PR-first adversarial review protocols, commit-pinned review ledgers, and bounded issue autonomy for Claude Code, Codex, and Gemini — synced from one upstream to your entire repository fleet.

$ npx activeloom init --harness claude --harness codex --harness gemini

✔ Claude Code ✔ OpenAI Codex ✔ Google Antigravity / Gemini ✔ Node 18.17+ & Python 3.9+ ✔ Apache 2.0

Native Agent Harnesses

Claude Code, Codex, and Gemini / Antigravity from one upstream.

10+

Production Skills

Hardened workflows for diagnosis, planning, task packets, and review.

SHA

Commit-Pinned Review Evidence

Review attestations name the exact head commit they read.

Apache 2.0

Permissive Open Source

No SaaS seat fees, no cloud lock-in. Host and distribute freely.

Live Protocol Simulator

The multi-agent workflow in action.

Click any agent, workspace, or tab to inspect live protocol evidence ↗

activeloom-session — 3 workspaces · 4 agents

protocol: converged sync: v2

Codex on api-gateway (PR #57 worktree) > feat/session-refresh

❯ /critique deep

[protocol] Draft PR #57 found. Review tier resolved: DEEP (Trigger: auth & session boundary)

[protocol] Preflight diff validation: 4 files changed, +142 lines, -18 lines.

[protocol] Running the deep review matrix...

✖ DEFECT DETECTED: Refresh token written to application logs

File: src/auth/refresh.ts#L142

Violation: raw request metadata, including the refresh token, is passed to logger.info(). Breaks the repo's no-secrets-in-logs invariant.

✔ Posted verified finding inline on PR #57 (commit: 4d2e1b).

✔ Automated fix diff prepared using redactSecrets() from shared package.

⠋ Awaiting second-opinion verification from Gemini... (42s elapsed)

PR #57 ledger — review evidence pinned to the head commit CONVERGED

<!-- local-review-roster:v2 author=claude reviewers=codex,gemini
     head=4d2e1b89ef01a23c7b89d2140f1a94e82b7c65d3 supersedes=none
     declaration-sha256=9f3c0d…e1a7 -->

✔ codex attested head 4d2e1b8 · 1 finding fixed · 0 open

✔ gemini attested head 4d2e1b8 · 0 findings

✔ verify-coverage: every declared reviewer attested the current head

A new commit moves the head and invalidates only the attestations that named the old one. The ledger is evidence, not a merge gate.

agent-loop on issue-88 worktree

❯ /agent-loop

[queue] Issue #88 allowlisted: "Fix allocator floor calculation on tiered accounts"

[worktree] Provisioned isolated worktree: /tmp/agent-loop-worktrees/issue-88 from origin/main

[repro] Generated regression test: tests/allocator-floor.spec.ts

[code] Modified packages/core/allocator.ts — complexity checked (24 logical lines)

✔ Test suite passed: 14 assertions clean in 840ms.

✔ Atomic conventional commit created: fix(allocator): clamp floor value to zero

✔ Opened draft PR #58. Local critique pass: 0 blockers.

✔ Issue #88 fixed on a draft PR. Waiting for maintainer review.

sync-engine v2 — Fleet propagation

❯ python3 scripts/sync-engine.py --upstream-repo ../activeloom --consumer-dir .

[sync] Source: loomantix/activeloom@sync-v2 (commit: 986a66b)

[sync] Updating upstream-managed skills and review workflows:

→ .agents/skills/critique/SKILL.md (updated)

→ .codex/REVIEW_WORKFLOW.md (synced)

→ .github/copilot-instructions.md (re-rendered with project variables)

★ Consumer-owned exceptions preserved without modification:

✔ .review/addendum.local.md (4 local security invariants protected)

✔ .codex/skills/agent-loop/agent-loop.config (local queue config preserved)

✔ Opened a propagation PR for review. Consumer-owned files untouched.

›

ctx: 4% (38k / 1M) esc to interrupt

Capabilities

Engineered for teams that treat agent code like production code.

ActiveLoom replaces fragile ad-hoc system prompts with structured engineering discipline, rigorous review ledgers, and deterministic fleet sync.

01 SKILLS & WORKFLOWS

Not chat prompts. Operational skills.

ActiveLoom skills encode strict operational procedures with validation checklists and deterministic outputs. From pre-code requirements grilling (grill) and deep troubleshooting (diagnosing-bugs) to task packets.

/diagnosing-bugs READY

Step 1: Reproduce defect in isolation

Step 2: Root cause analysis (no guessing)

Step 3: Smallest meaningful fix + regression test

02 MULTI-ENGINE PARITY

Claude, Codex, and Gemini in sync.

Don't maintain three divergent prompt trees. Shared skills are authored once in prompts/ and rendered into .claude/, .codex/, and .agents/ (Gemini / Antigravity). Review prompts are tuned per engine on purpose.

One upstream definition 3 native outputs

Claude Code

OpenAI Codex

Gemini / Agy

03 REVIEW PROTOCOL

Adversarial review. Proof before edit.

No rubber-stamping. Reviewers must inspect draft PR diffs and post verified findings inline to GitHub before touching any code. Automated tier gating resolves Lean (fast critical check) or Deep (exhaustive matrix).

Review Tier Gate: Deep Tier

Trigger: auth & crypto boundaries

✓ Findings posted inline before code fix

04 REVIEW LEDGER

Review evidence pinned to the commit.

Every review pass is recorded on the draft PR as ledger markers carrying sha256 digests and the exact head commit. Push a new commit and attestations for the old head stop counting.

PR ledger · verify-coverage HEAD MATCHED

head=4d2e1b89ef01a23c7b89...

✓ Attested by Codex + Gemini reviewers

05 BOUNDED AUTONOMY

Autonomous queues with guardrails.

Let agents work your GitHub backlog safely with agent-loop. Strict issue allowlists, one isolated Git worktree per issue (/tmp/agent-loop-worktrees by default), test-driven iteration, and automatic bail-outs when blocked.

CLI Tape Demo: LIVE

06 FLEET DISTRIBUTION

Sync every repo without losing local rules.

ActiveLoom's sync engine propagates skills and review updates via PRs (sync-v2) while strictly protecting consumer-owned security rules in .review/addendum.local.md.

Upstream: activeloom@sync-v2 PR opened

Syncs: Core skills, review workflows

★ Local addenda & project rules strictly preserved

Hard Boundaries

What ActiveLoom deliberately does NOT do.

Software engineering tools earn trust through explicit non-goals. We don't lock you into a cloud, hide the prompts, or replace human review.

✔ What ActiveLoom delivers

✔ Standardized engineering skills: Planning, debugging, and review workflows that any team agent can invoke.
✔ PR-first adversarial review: Verified findings posted inline to GitHub before code modifications occur.
✔ Commit-pinned review ledgers: Review evidence recorded on the PR and bound to the exact head commit it covers.
✔ Upstream fleet sync: Keep every repository updated via proposed PRs without clobbering repo-local rules.

✖ What we deliberately avoid

✖ No proprietary agent runtime: Bring your own Claude Code, OpenAI Codex, or Gemini CLI. No vendor lock-in.
✖ No monthly SaaS seat tax: No hosted workspace fees, no credit tokens. Apache 2.0 open source, with a zero-dependency CLI.
✖ No silent auto-merges: agent-loop opens draft PRs and the review runner never marks ready or merges. Shipping stays a human decision.
✖ No clobbering of local rules: Upstream prompt improvements never overwrite your repository's custom compliance invariants.

Adoption Model

From personal trial to fleet sync.

Adopt gradually. Start by running a single skill in your personal shell, then scale up to scheduled multi-repo propagation.

Tier 0

Personal Evaluation

Try a single skill in your home directory without touching any shared repository files or configurations.

$ activeloom add critique

Tier 1

Repository Setup

Scaffold harness directories (.codex/, .claude/) and shared project instructions for your team.

$ activeloom init

Recommended

Tier 2

Scheduled Sync

Automated GitHub Actions workflow proposing upstream skill improvements via pull requests using built-in GITHUB_TOKEN.

$ activeloom init --sync

Tier 3

App-Backed Fleet

Scheduled sync through a GitHub App, so propagation PRs land with GitHub-signed commits.

$ activeloom init --sync --app

Ready to evaluate?

Give your agents real discipline today.

One command to initialize your repository. Your workflow stays yours. Your agents' reviews now leave evidence pinned to the commit they read.