AI AGENT WORK RECORDER · OPEN SOURCE

Work moves forward.
The trail stays.

Your AI agent does not need another flight plan. Let it work. flightwake leaves state, decisions and evidence at the moments that matter, so the next session can pick up safely.

Install flightwake See the workflow

Published on npm: 0.16.0Easier setup · Claude Code mod 0.2.0

SESSION 01 → 02 → 03

REPO / FLIGHT RECORDER↗

01 /fw-coldstart

Read STATE + latest record

02 Build, test, review

Keep decisions and traps as they arise

03 /fw-record

The next agent starts here →

●STATE.mdREADY

01 / THE GAP

Models have context limits. Projects do not.

A session ends, but the work in the repo continues. The expensive part is making the next agent guess what changed, what remains unverified, and why a path was chosen.

STATE

Where we are

Current state, honest health, and the next entry point.

RECORDS

What happened

Changes and verification evidence at each meaningful wrap-up.

DECISIONS + TRAPS

Why, and where we tripped

Reasons behind choices and surprising failures worth remembering.

Records follow work. They do not lead it.

One less plan to maintain. One more trail you can trust.

02 / ONE CONTINUOUS LOOP

You say “start.” Events trigger the rest.

01 / START

Read state, take over safely

A new session reads STATE and the latest record. Without evidence, health never gets guessed green.

/fw-coldstart

02 / WORK

Build, and leave clues

When a decision rules out alternatives, record why. When a surprising trap appears, write it down immediately.

DECISIONS + TRAPS

03 / WRAP UP

Pass the evidence on

Meaningful work ends with tests, risks and next steps in a record. STATE stays short and honest.

/fw-record

04 / CONTINUE

Switch models, keep context

Claude Code, Codex and Gemini CLI read the same repository trail, independent of a chat window.

git + Markdown

START

CONTINUE ↗

03 / 0.16.0 / ACTUAL OUTPUT

What do you actually see inside Claude Code?

Version 0.16.0 brings optional add-ons into guided setup and adds cost and session status to Claude Code mod 0.2.0. These are excerpts from real commands in an isolated test repo. The terminal frames are laid out for this site, and temporary paths are redacted.

FLIGHTWAKE · FIELD CAPTURE01 / 03

npx flightwake setup

The core (records, the four skills, the wrap-up hook) is always installed. Optional add-ons — * = recommended:
  [*] 1) Bottom gauge in Claude Code (health / STATE lag / context usage)
  [*] 2) Claude Code mod — inside Claude Code: a status row above the prompt, STATE loaded when a session starts, a log of this session's changes and test runs (/fw-log), a hint when you touch something a known trap mentions, and an optional role guard.
       Needs Claude Code 2.1.287 or later. It loads in a Claude Code session started from the repo root, after you accept the folder trust prompt — a session that is already open needs a restart.
  [ ] 3) Team roles (installs only the fw-roles skill; your agent recommends roles later, nothing applied yet)
Enter = the * ones · or type the numbers you want (e.g. 1 2) · 0 = none:

01 / SETUP

See before installation

Setup lists the bottom gauge and mod separately, marks recommendations, and previews write paths before confirmation.

FLIGHTWAKE · FIELD CAPTURE02 / 03

/fw-mod

flightwake-mod v0.2.0

STATE at session start (stateInject): on — STATE is not initialized yet, so only a "run /fw-coldstart" note is injected
band above the prompt (band): on — hiding the fields the bottom gauge already shows; only the one-time 80% context toast remains
session flight log (recorder): on — this session: 0 file(s) by tools, 0 via shell (inferred), 0 test run(s), 0 commit(s) — /fw-log prints them
trap tripwire (tripwire): idle — no active TRAPS entry has paths or commands
role guard (roleGuard): off — opt-in
tokens and cost (spend): on — this session: tokens unknown · API-equivalent ≈ $0.00 — /fw-cost prints the breakdown
session fleet (fleet): on — this session is reported as idle in [isolated test sessions] (heartbeat 30 s; never prompt or command text)

Read-only: /fw-mod changes nothing.

02 / CLAUDE CODE

Seven features, visible state

Run /fw-mod in Claude Code to see which features are on, idle or off, and why. With the bottom gauge active, the mod band avoids duplicate fields.

FLIGHTWAKE · FIELD CAPTURE03 / 03

npx flightwake fleet --lang=en

flightwake fleet — 2 live · ⛔ 1 blocked — [isolated test sessions] (last 7 days)
⛔ blocked  demo-review              demo-re…    21s  permission · Bash
● working   demo-build               demo-bu…    14s  Read src/app.ts

03 / FLEET

See who is waiting

Fleet puts blocked sessions first. This real CLI output uses controlled test session data.

Setup ran through the test driver with supplied answers; /fw-mod ran in an interactive Claude Code session; the released CLI read Fleet test fixtures. Excerpts are reformatted with temporary paths redacted. No live user data is shown. Some permission-dialog and crash-resume cases still need real-session verification.

Read the full mod docs ↗View the 0.16.0 changelog ↗

04 / FIELD NOTES / ORCA × FLIGHTWAKE

When agents actually work together, who remembers the whole story?

This project uses flightwake on itself. Orca keeps parallel agent conversations and reviews visible in tabs. flightwake returns accepted findings, tests and open questions to the repository.

CASE STUDY · FLIGHTWAKE REPOSITORY 2026.10

ORCA / VISIBLE WORK

Parallel tasks and reviews

Agents inspected Fleet, installation and edge cases in parallel. The lead integrated their findings and reran the checks.

FLIGHTWAKE / SHARED MEMORY

One writer, one source of truth

Review agents did not each rewrite STATE. The lead recorded accepted results, letting the next session start from one place.

REAL RESULT / 2026.10

Evidence, with unfinished work

The record reports 46 smoke sections and 465 mod tests passing, while naming Fleet scenarios still awaiting real-session checks.

Read the actual flight record ↗ORCA × FLIGHTWAKE

* Orca is optional. Fleet shipped in 0.16.0; some real-session scenarios still need verification.

ANOTHER FIELD NOTE / ANONYMOUS PRODUCT TEAM

Two repos. Four roles. One traceable line of work.

One repo holds product decisions; another holds engineering work. Orca gives agents visible places to collaborate. flightwake leaves the current state, handoffs and verification boundary in both repos.

01 / PRODUCT LEAD

Define and delegate

A product agent sets the goal and acceptance criteria, approves a limited plan, then assigns implementation to engineering.

02 / TECH LEAD

Challenge the risks

A technical agent reviews the approach and its boundaries before the code is treated as settled.

03 / ENGINEER

Build and self-test

An engineering agent implements in the product repo, runs local tests and type checks, and records changes still awaiting a freeze.

04 / INDEPENDENT REVIEW

Verify before acceptance

Review and security roles check the frozen implementation independently. Until then, the work remains in progress.

Field snapshot · 2026.10: implementation in progress; independent review incomplete. No claim of product acceptance or launch.

05 / WHAT YOU GET

Light enough to keep using.

Markdown in git

No separate database. STATE, records, decisions and traps live in the repo.

Three coding agents

Claude Code, Codex and Gemini CLI get their own skills and wrap-up hooks over one shared work record.

Event-triggered discipline

Start, decisions, surprises and wrap-up each trigger a small action, without turning work into a form.

See before you install

Guided setup previews the language, agents, optional add-ons and every path it will write.

Read-only health check

doctor checks installation structure; STATE honestly marks verified and unverified work.

Add collaboration as needed

The status gauge, Orca instructions, team roles and Claude Code mod are optional.

06 / START HERE

A few questions, inside your repo.

Requires Git and Node.js 18 or newer. setup asks for language, agents, optional add-ons and repo type, then previews every write before installation.

Read the full installation guide ↗

01 / Guided setup

$ cd your-repo
$ npx flightwake setup

02 / Automation / noninteractive

$ npx flightwake init --lang=en --agents=claude,codex,gemini

Open an agent session in the repo and run its coldstart skill. From then on, let the work trigger the records.

Already installed? Run npx flightwake update. To add the Claude Code mod, use npx flightwake update --mod.

07 / FAQ

Frequently asked questions

Does flightwake plan my work?

No. It is a work-recording framework. Agents work directly, preserving enough context at cold start, decisions, surprises and wrap-up for the next session to take over.

Do I need Orca or Claude Code?

No. flightwake supports Claude Code, Codex and Gemini CLI. Orca collaboration and the Claude Code mod are optional.

Where is the data stored? Is a cloud account required?

Core data is Markdown in the repo, including .flightwake/STATE.md and records/. No flightwake cloud account or runtime package dependencies are required.

Can I install 0.16.0 now?

Yes. Version 0.16.0 is published on npm with guided setup improvements and the optional Claude Code mod 0.2.0. Some Fleet real-session scenarios remain unverified.

How do I update an existing installation?

Run npx flightwake update. It keeps your existing options and .flightwake user data; optional add-ons are not installed automatically.

Work moves forward.The trail stays.