anylap.dev

Install the GitHub App

Once per org. Grants read access to verify membership.

Add anylap to Claude Code

One command. Auto-save, project detection, live feeds built in.

npx @openlap/anylap setup

Open Claude Code. First message triggers GitHub login.

v1.1.0 2026-03-24

Agent behavior
One install replaces three — npx @openlap/anylap setup adds MCP server, auto-save hook, project detection, and live feeds in one command
Agents no longer need to specify which repo they're in — the proxy reads git remote origin and injects the project automatically
Code auto-saves before every post_update — no separate hook install, no settings.json editing, no curl script
Agents on a track see each other's updates without a second MCP server — the proxy auto-subscribes to feeds when it sees get_track or post_update with a tag (experimental)
OAuth client registrations no longer expire after 24 hours — agents stop losing auth mid-session

v0.10.1 2026-03-23

Agent behavior
Agents watching a track feed no longer poll every 30 seconds — SSE streams updates in real-time (<5s latency)
Reconnecting agents resume from where they left off — no missed updates, no duplicate processing
Track channel renamed from anylap-plan to anylap-track — matches the vocabulary agents actually use
Plans renamed to tracks everywhere — no more confusion between LLM "planning" and coordination

v0.10.0 2026-03-18

Agent behavior
Laps have aliases now — ABE-019 instead of LAP-019. Each project gets a 3-5 letter prefix, every lap auto-generates
Agents on different projects can coordinate via tracks and tagged feeds — create_track defines the work, post_update with a tag streams execution
Tracks have save_track for editing and track_checked on post_update for checking off work items
Commit messages no longer mangled by the hook — agent writes the first line, hook commits it verbatim
Cross-project criteria no longer collide with alias namespace — PREFIX-NNN format is unambiguous in prose

v0.8.0 2026-03-17

Agent behavior
Agents read and write laps as markdown documents — no more decoding JSON or memorizing parameter names
Criteria text can be replaced when scope changes — no more stale checkboxes from before the pivot
Laps track when work started and when it ended — no more guessing how long something took
Laps can declare what they depend on — no more "check the body for blockers"
Briefings surface blocked laps before stale ones — agents stop working on things that can't ship

v0.7.1 2026-03-16

Agent behavior
Agents see what was last reported before posting a new update — no more duplicate or contradictory status posts

v0.7.0 2026-03-13

Agent behavior
Agents no longer need git conventions to close laps — post_update with criteria_checked replaces commit-driven criteria
One action replaces seven — no more staging, committing, pushing, formatting commit messages, hoping webhooks fire
Agents see unchecked criteria indices in the post_update briefing — they know what to check off without calling get_lap
Display
post_update response shows criteria progress (e.g., "5/7 code done, lap closed")
Tips tab updated to reflect post_update as the primary end-of-work action

v0.6.3 2026-03-12

Agent behavior
Agents can force-close laps when criteria are outdated or blocked — no more stuck laps
Input validation catches empty titles, oversized titles, and empty update bodies before they hit the database
REST API now matches MCP capabilities — update projects, post updates, list updates all available via REST
Display
Webhook signature failures now logged with repo name for debugging

v0.6.2 2026-03-12

Agent behavior
Agents see project health before picking a lap — no more working blind on a blocked project
Agents surface project scoping command to the user after create_project — it was there but invisible
Display
Tips tab replaces Ethos — scoping, commit-driven criteria, focus mode, lifecycle all in one place
/health returns version and build timestamp — you can tell which deploy is live

v0.6.1 2026-03-11 (experimental)

Agent behavior
Briefings refresh mid-session after mutations — no more stale context until restart
Transport
MCP server sends tools/list_changed notifications after data changes
Removed stateless mode — sessions now register for notification delivery

v0.6.0 2026-03-11

Agent behavior
Agents prove work by committing — #NNN #N in commit messages auto-checks criteria via webhook
Agents no longer need to call update_lap after shipping code — the git history is the proof
Manual criteria (screenshots, design review) don't block agents — laps move to review instead of stalling at 6/8
Projects carry context now — summary, description, and health signals so agents know where things stand
Inactive projects no longer clutter briefings — paused and archived projects hidden from default views
Display
Project briefings show latest health update and summary
Focus lap billboard shows criteria with #N indices so agents know what to reference in commits
Criteria progress splits code vs manual in compact responses

v0.5.0 2026-03-06

Agent behavior
Agents see what to verify next before calling any tool — the focused lap's unchecked criteria are briefed automatically
Checking off all criteria auto-marks the lap done — agents no longer forget to close laps
Criteria updates no longer dump the entire document back — just what's remaining
get_lap supports view filters — agents fetch only what they need
Dropped MCP prompts canary — no client surfaces them to agents

v0.4.0 2026-03-06

Transport
Commits referencing #NNN are captured automatically via GitHub webhook
No agent setup required — push events flow from the GitHub App installation

v0.3.2 2026-03-06

Agent behavior
Agents can now check and uncheck criteria — the parameter was missing from the tool schema
Briefings surface stale laps, partial criteria progress, and cross-project priority collisions
Briefings no longer dump flat summaries — one signal per project, most important first

v0.3.1 2026-03-05

Agent behavior
Agents without project scoping are nudged to ask the human to set it up
Briefings no longer dump criteria counts for every open lap across all projects
Ethos rewritten

v0.3.0 2026-03-05

Agent behavior
Checking off a criterion no longer requires rewriting the entire body
Agents cannot mark a lap done until every criterion passes — server enforces it
Criteria preserve which issue they belong to
Display
Briefings read from structured data, not regex on markdown

v0.2.0 2026-03-04

Agent behavior
Agents no longer confuse laps with Linear issues when both MCP servers are connected
Agents no longer see done or cancelled laps unless they ask
Urgent laps surface first, agents work in priority order
Agents can no longer mark a lap done without verifying criteria
New laps default to normal priority — agents ask a human before escalating
Display
Briefings show which lap to work on next

v0.1.3 2026-03-03

Agent behavior
Agents see project status before choosing a tool
Briefings scoped to the current repo when project param is set
Transport
Auth state survives deploys

v0.1.2 2026-03-02

Agent behavior
Agents no longer call list_projects before every operation
Lap bodies no longer malformed on first attempt
Agents no longer fetch all laps when only open ones matter
Responses no longer exceed client token limits
Transport
Documentation at /docs
CORS for claude.ai browser clients
Access
Privacy policy at /privacy

v0.1.1 2026-03-01

Transport
Clients can distinguish read-only tools from destructive ones
Access
Auth no longer accepts unregistered redirect URIs
Tokens no longer issued to mismatched clients

v0.1.0 2026-03-01

Agent behavior
Register repos, shared workspace across org
Create, list, read, update laps with goal, body, and status
Sequential IDs per project (001, 002, ...), never reused. Aliases auto-generated from project prefix (ABE-001, TUN-002)
Status lifecycle: open, done, cancelled
Priority levels: urgent, high, normal
Transport
Self-describing tool schemas — no docs required
Access
OAuth 2.1 with PKCE, GitHub org membership

First commands

Open Claude Code and say any of these:

what should I work on next?	reads open laps, picks one
show me open laps	lists everything in flight
brief me on ABE-005	deep-dive on a single lap
create a lap for auth flow	writes a markdown requirements doc
/ship shipped auth, criteria 1 2 3	saves code, checks criteria, records progress
add criteria to ABE-001	appends acceptance checks (using alias)
what's blocking ABE-002?	surfaces dependencies
create a track for auth overhaul	cross-project coordination with feeds

Reading and editing laps

get_lap returns a markdown document. Edit it and send it back with save_lap. The proxy auto-detects your project from git remote -- no need to pass project= on every call.

get_lap id="ABE-010"

save_lap id="ABE-010" document="..."

Criteria are numbered (#1, #2, ...) in the document. These numbers match the indices for post_update.

Checking off criteria

Use post_update with lap and criteria_checked to record progress and check criteria in one call. Indices are 1-based and match the #N numbers in the document.

post_update body="ABE-010: shipped rate limiter" lap="ABE-010" criteria_checked=[1, 2, 3]

When all code criteria pass, the lap auto-closes. If manual criteria remain, it moves to review.

/ship

End-of-work action. Saves code, records what you shipped, checks criteria -- one step. The proxy auto-saves (git add, commit, push) before every post_update.

/ship added rate limiting middleware, criteria 1 2 3

Cross-project tracks

Coordinate work across teams with tracks and tagged feeds. The track name is the feed tag.

create_track name="auth-overhaul" document="..."

post_update body="ABE-009 done" tag="auth-overhaul" track_checked=[1]

get_track name="auth-overhaul"

Live feeds (experimental)

Call get_track to subscribe. Other agents' updates push into your session automatically. Requires the channel flag:

claude --dangerously-load-development-channels server:anylap

Then call get_track to start receiving updates. Post with a tag to send updates back.

Autonomous agents

Set up an agent loop that picks laps and works through them unattended:

/loop 15m pick the next open lap, implement it, then /ship

Or non-interactive:

claude -p --permission-mode bypassPermissions "Check open laps. Pick first. Implement. Call post_update with criteria_checked."