Turn a one-line idea into a branch with tested, reviewed, committed code. The brainstorm-to-commit pipeline for Claude Code.
> /plugin marketplace add LucasDuys/forge> /plugin install forge@forge-marketplace
A Claude Code plugin that turns a brief into a cinematic launch video, an investor pitch deck, or a feature walkthrough — with live FLIP morph transitions, audio bus, scene SFX, and a dev scrubber wired the first time.
@lucasduysFAQ
forge is a Claude Code plugin with 10 hand-picked skills for development work, indexed on Flowy. Install it with the command on its page. It includes backpropagation, brainstorming, caveman-internal. Its skills do not fire on their own yet. Request auto-invocation to have Flowy route them as you prompt. Free and open source.
Repo: LucasDuys/forge
You start a feature in Claude Code. You write the prompt. It writes the code. You review it. You re-prompt. It tries again. It loses context. You re-explain. You watch the "context: 87%" warning crawl up. You restart. You re-explain again. Three hours in, half a feature done, and you are the one keeping the whole thing from falling apart.
You are the project manager. You are the state machine. You are the glue.
Forge replaces you as the glue. You describe what you want in one line. Forge writes the spec, plans the tasks, runs them in parallel git worktrees with TDD, reviews the code, verifies it against the acceptance criteria, and commits atomically. You read the diffs in the morning.
A native Claude Code plugin that turns one-line ideas into reviewed, tested, committed code through a five-phase autonomous loop:
State lives on disk in .forge/, not in a conversation window. Crashes, context resets, and OOMs are recoverable because the state machine restarts from disk, not memory.
Two minutes. Requires Claude Code v1.0.33+. Zero npm install for the solo path.
claude plugin marketplace add LucasDuys/forge
claude plugin install forge@forge-marketplace
That's all you need for single-user runs. Multiplayer mode (/forge:collaborate) optionally adds Ably for sub-second cross-machine coordination — see docs/collaborate.md.
Three commands. One autonomous loop. One squash-merge.
/forge brainstorm "add rate limiting to /api/search with per-user quotas"
/forge plan
/forge execute --autonomy full
Walk away. This is what you actually see while it runs.
$ /forge brainstorm "add rate limiting to /api/search with per-user quotas"
[forge-speccer] generating spec from idea...
spec written: .forge/specs/spec-rate-limiting.md
R001 per-user quotas, configurable per tier (free / pro / enterprise)
R002 sliding window counters (1 minute, 1 hour, 1 day)
R003 429 response with Retry-After header
R004 bypass for admin tokens
R005 redis-backed counters with atomic increment
R006 structured logs for rate-limit events
R007 integration test against /api/search
$ /forge plan
[forge-planner] decomposing into task DAG...
8 tasks across 3 tiers (depth: standard)
T001 add redis client + connection pool [haiku, quick]
T002 implement sliding window counter [sonnet, standard]
T003 build rate-limit middleware [sonnet, standard]
T004 wire middleware to /api/search route [haiku, quick]
T005 add 429 response with Retry-After [haiku, quick]
T006 admin token bypass [haiku, quick]
T007 structured logging [haiku, quick]
T008 integration test [sonnet, standard]
deps: T001 T002 T003 T004 T005 T006 T007
$ /forge execute --autonomy full
══ FORGE iteration 3/100 ══════════════════════════════════ phase: executing ══
Task T002 [in_progress] @ tests_written → tests_passing
Tasks [████████░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░] 1/8 (12%)
Tokens 47k in / 12k out / 23k cached budget 47k/500k (9%)
Per-task 8k/15k tok (53%)
Lock alive pid 18432, 4s ago restarts 0/10
──────────────────────────────────────────────────────────────────────
[14:02:48] T001 PASS 4 lines, 1 commit, budget 1820/5000
[14:02:48] T002 T003 dispatched in parallel (disjoint files)
[14:06:01] T003 PASS 62 lines, 8 tests, budget 13880/15000
[14:08:27] tier 2 complete, squash-merged 6 worktrees
[14:14:18] forge-verifier: existence > substantive > wired > runtime
[14:14:18] verifier PASS all 7 requirements satisfied
[14:14:18] <promise>FORGE_COMPLETE</promise>
8 tasks. 12 minutes. 218 lines. 9 commits squash-merged to main.
session budget: 47200 / 500000 used. lock released.
You read the diffs. You merge the branch. You move on.
The pipeline is strictly sequential, enforced programmatically: brainstorm → plan → execute. You cannot skip brainstorming, skip planning, or bypass the approval gate. The spec is the contract. Every acceptance criterion has an R-number; every task maps to at least one R-number; the verifier checks R-numbers, not checklists.
| Action | gated (default) | full |
|---|---|---|
| Write spec from your one-line idea | automatic (asks you questions during Q&A) | automatic |
| Decompose spec into tasks | automatic | automatic |
| Write code + tests for each task | automatic | automatic |
| Run tests, review, verify each task | automatic | automatic |
| Squash-merge passing tasks to the working branch | automatic | automatic |
| Install a new dependency not in the manifest | pauses and asks | assumes prior consent, installs |
| Hit a paid API (Stripe, OpenAI beyond Claude) | pauses and asks | assumes prior consent, calls |
| Push to a remote | pauses and asks | pauses and asks (both modes require explicit approval) |
| Run destructive git ops (force push, reset --hard) | refuses unless the spec explicitly requests | refuses unless the spec explicitly requests |
| Propose a spec update when tests hit a gap | automatic (proposal in .forge/backprop-log.md) | automatic, applied immediately on high-confidence gaps |
The headline difference: full mode assumes you already authorized the side-effect class when you ran /forge:execute --autonomy full. It still refuses destructive git ops and still pauses before pushing.
<promise>FORGE_COMPLETE</promise> is a structural gate: tasks done, tests green, reviewer satisfied, verifier satisfied. A feature that passes all four can still look broken in the browser (blurred canvas, empty panel, wrong state after a click) because unit tests don't render pixels.
When that happens:
/forge:backprop "<what-is-wrong>". Backprop traces the bug to the R-number whose acceptance criteria should have caught it, proposes a tightened criterion, and generates a regression test that would have failed against the shipped code./login and the session cookie is cleared"), then rerun /forge:execute.For visual ACs in 0.3+, the verifier opts into a perceptual gate: [visual] path=/login occluded_check=true selector="#login-form" runs through Playwright with deterministic readiness (document.fonts.ready + animations-disabled + 2× rAF) and an elementFromPoint-based occlusion probe so a target hidden behind a modal fails the AC rather than passing silently. See docs/visual-verification.md.
.forge/resume.md and stops cleanly. /forge:resume picks up where it died, no re-explaining. (docs/budgets.md)PostToolUse filter trims long Bash output (npm install, git diff, tsc, find, curl) to head + warnings + tail; a 120 s read-tool cache collapses repeat file reads; caveman compression strips fillers from agent-to-agent handoff text; per-role max_tokens caps (2k–16k) bound every Claude turn. Measured against a real 13.8k-LOC project (Stacklink/teambrain) in two ways:
git diff consumed 59,600 tokens with the raw 57 KB input versus 42,402 tokens with the filtered 20 KB input — 28.9% fewer real Claude tokens. The byte reduction (65%) doesn't translate 1:1 because the model's reasoning + output overhead is roughly constant; only the input-context portion shrinks. Real-run savings are highly task-dependent: a workload with no filterable Bash, no repeat reads, and no verbose handoffs sees ~0% reduction. (docs/budgets.md)/forge:resume reconstructs state and continues. (docs/recovery.md)flowchart LR
User([You: one-line idea]) --> Bs["/forge brainstorm"]
Bs --> Spec[".forge/specs/spec-{domain}.md<br/>R001…R0NN + acceptance criteria"]
Spec --> Plan["/forge plan"]
Plan --> Frontier[".forge/plans/{spec}-frontier.md<br/>tier 1 ┃ tier 2 ┃ tier 3<br/>dependency DAG"]
Frontier --> Exec["/forge execute"]
Exec --> Loop{"autonomous<br/>loop"}
Loop -->|all done| Done([squash-merge to main<br/>FORGE_COMPLETE])
Loop -.->|read-only| Watch["/forge watch<br/>live TUI dashboard"]
Loop -.->|read-only| Headless["/forge status --json<br/>headless query"]
Crash[crash / context reset] -.->|/forge resume| Loop
classDef cmd fill:#e3f2fd,stroke:#1565c0,color:#0d47a1
classDef state fill:#fff3e0,stroke:#e65100,color:#bf360c
classDef ui fill:#e0f7fa,stroke:#006064,color:#004d40
classDef done fill:#c8e6c9,stroke:#1b5e20,color:#0d2818
class Bs,Plan,Exec,Loop cmd
class Spec,Frontier state
class Watch,Headless ui
class Done,User done
class Crash state
The state machine drives everything. The Stop hook fires routeDecision() after every Claude turn and picks the next phase based on .forge/state.md. Seven hooks fire on every executor tool call to cap tokens, condense test output, cache repeat reads, track progress, and trigger auto-backprop on test failure. Detailed walkthroughs of the execute loop, hooks pipeline, recovery layer, and team mode live in docs/architecture.md.
Start with one of these depending on what you need:
node scripts/run-tests.cjsSee CONTRIBUTING.md.
.claude-plugin/
marketplace.json
plugin.json
.github/
ISSUE_TEMPLATE/
bug_report.yml
feature_request.yml
PULL_REQUEST_TEMPLATE.md
.gitignore
agents/
AGENTS.md
forge-complexity.md
forge-executor.md
forge-planner.md
forge-researcher.md
forge-reviewer.md
forge-speccer-validator.md
forge-speccer.md
forge-verifier.md
forge-visual-verifier.md
CLAUDE.md
commands/
backprop.md
brainstorm.md
collaborate.md
execute.md
help.md
plan.md
resume.md
review-branch.md
setup-tools.md
skills-audit.md
status.md
update.md
watch.md
CONTRIBUTING.md
docs/
agents.md
architecture.md
audit/
CLAIM_VS_REALITY.md
mock-verify-evidence/
halo/
after.png
before.png
README.md
synthesis/
after.png
before.png
zoomOut/
after.png
before.png
research/
streaming-dag.md
backpropagation.md
benchmarks/
caveman-integration.md
streaming-dag-vs-task-dag.md
worktree-overhead.md
budgets.md
caveman.md
collaborate.md
commands.md
comparison.md
configuration.md
dashboard.md
forge-design-visual.html
headless.md
index.html
mechanics/
caveman-details.md
commands-reference.md
subsystem-reference.md
token-savings.md
recovery.md
superpowers/
plans/
2026-03-17-forge-implementation.md
collab-fix-frontier.md
forge-v03-gaps-frontier.md
frontier.md
mock-and-visual-verify-frontier.md
specs/
2026-03-17-forge-design.md
2026-04-09-tui-dashboard.md
spec-collab-fix.md
spec-collab-frontier.md
spec-collab.md
spec-forge-v03-gaps.md
spec-mock-and-visual-verify.md
testing.md
verification.md
worktrees.md
hooks/
auto-backprop.js
hooks.json
output-filter.js
progress-tracker.js
stop-hook.sh
test-output-filter.js
token-monitor.sh
tool-cache-store.js
tool-cache.js
LICENSE
mock-projects/
blurry-graph/
.forge/
config.json
specs/
001-readable-graph.md
.gitignore
demo.sh
index.html
package.json
README.md
src/
App.tsx
config.ts
main.tsx
tsconfig.json
vite.config.ts
package.json
README.md
references/
backprop-patterns.md
checkpoint-schema.md
complexity-heuristics.md
config-schema.md
forge-directories.md
headless-status-schema.md
model-routing.md
multi-repo.md
review-protocol.md
state-machine.md
token-profiles.md
scripts/
bench-caveman-agents.cjs
bench-caveman-writes.cjs
bench-worktree-overhead.cjs
forge-budget.cjs
forge-collab-schema.json
forge-collab-scorer.cjs
forge-collab.cjs
forge-dev-server.cjs
forge-head-cache.cjs
forge-research-aggregator.cjs
forge-router.cjs
forge-runner.sh
forge-skills-audit.cjs
forge-speccer-validator.cjs
forge-status-block.cjs
forge-streaming-dag.cjs
forge-token-bucket.cjs
forge-tools.cjs
forge-tui-attach.cjs
forge-tui.cjs
forge-update.cjs
forge-usage-capture.cjs
forge-wizard.cjs
run-tests.cjs
setup.sh
skills/
backpropagation/
SKILL.md
brainstorming/
SKILL.md
test.md
caveman-internal/
references/
budget-thresholds.md
SKILL.md
collaborating/
SKILL.md
design-system/
SKILL.md
executing/
SKILL.md
graphify-integration/
SKILL.md
karpathy-guardrails/
SKILL.md
planning/
SKILL.md
reviewing/
SKILL.md
templates/
artifact.json
backprop-report.md
collab-gitignore
config.json
gitignore.forge
plan.md
resume.md
spec.md
state.md
summary.md
tests/
_helper.cjs
auto-backprop.test.cjs
brainstorm-check-design.test.cjs
budget.test.cjs
caveman-whitelist.test.cjs
checkpoints.test.cjs
collab-claim-race.test.cjs
collab-disconnect-race.test.cjs
collab-polling-branch.test.cjs
completion-gates.test.cjs
detect-contention.test.cjs
dev-server-lifecycle.test.cjs
discover.test.cjs
fixtures/
bucket-payloads.json
synthetic-spec.md
tool-cache-v1-snapshot.json
transcript-sample.jsonl
forensic-recovery-test.cjs
forge-collab-ably-cas.test.cjs
forge-collab-bounded-queue.test.cjs
forge-collab-enabled-marker.test.cjs
forge-collab-gitignore.test.cjs
forge-collab-integration.test.cjs
forge-collab-polling-real.test.cjs
forge-collab-scorer.test.cjs
forge-collab-target-filter.test.cjs
forge-collab-wire.test.cjs
forge-collab.test.cjs
forge-e2e-smoke.test.cjs
forge-handoff-plumbing.test.cjs
forge-hands-off-verify.test.cjs
forge-head-cache.test.cjs
forge-help-killswitch.test.cjs
forge-regression-budget.test.cjs
forge-router-effort.test.cjs
forge-token-bucket.test.cjs
forge-token-opt-parity.test.cjs
forge-tools-cache-stats.test.cjs
forge-tools.test.cjs
forge-tui/
forge-tui.test.cjs
backoff-test.cjs
bucket-bar-test.cjs
bucket-bar-v1.snap
bucket-bar-v2.snap
fallback-test.cjs
fixture-stream.jsonl
parser-test.cjs
reconciler-test.cjs
render-test.cjs
run.cjs
snapshot-render.txt
v21-integration-test.cjs
forge-update.test.cjs
forge-usage-capture.test.cjs
forge-wizard.test.cjs
frontier.test.cjs
frontmatter-merge.test.cjs
headless.test.cjs
ledger-v2-migration.test.cjs
list-cited-docs.test.cjs
locks.test.cjs
mock-demo-evidence.test.cjs
mock-e2e-fix-run.test.cjs
mock-isolation.test.cjs
output-filter.test.cjs
playwright-unavailable.test.cjs
researcher-dispatch.test.cjs
route.test.cjs
run-tests-budget-w2.test.cjs
run-tests-perf.test.cjs
setup-state-guard.test.cjs
setup.test.cjs
skills-audit.test.cjs
spec-path-validation.test.cjs
speccer-validator.test.cjs
state.test.cjs
status-block.test.cjs
status-tokens.test.cjs
streaming-dag-forge-landing-shape.test.cjs
streaming-dag.test.cjs
task-classify.test.cjs
token-cache-smoke.test.cjs
token-instrumentation-smoke.test.cjs
tool-cache-head-pinning.test.cjs
tool-cache-kill-switch.test.cjs
tool-cache-mtime.test.cjs
tool-cache-patterns.test.cjs
tool-cache-store.test.cjs
tool-cache-ttl.test.cjs
transcript-append-cli.test.cjs
transcripts.test.cjs
tui-auto-attach.test.cjs
verify-structural-acs.test.cjs
visual-verifier-crlf.test.cjs
visual-verifier-occlusion.test.cjs
visual-verifier-readiness.test.cjs
visual-verifier.test.cjs
wave4-regression-budget.test.cjs
worktrees.test.cjs© 2026 Flowy · Free and open source
Built for Claude Code · Not affiliated with Anthropic
0.3+fonts.readyelementFromPoint/forge:status --json into Prometheus or a cron job. (docs/headless.md)