LedgerInfra
Menu
Start free

QUICKSTART

Use your first proposal in the next AI task.

Download the tools and a sample design proposal. Publish it privately, review a revision, then ask a new AI task to retrieve and use that exact source.

Start free

SET UP ONCE

Get the tools before you sign up.

Use Codex or Claude Code on macOS or Ubuntu / Debian. Install the CLI for your computer, check its SHA-256, then copy the Skill installation prompt into a project. Keep the same project for this walkthrough.

Run led --version to check the CLI. Read the public installation instructions if your agent cannot find the Skill.

Create a free account when you are ready to publish. Google uses its verified email state; email/password accounts need email verification. Your agent will start led login; compare the browser and terminal codes and approve sign-in yourself.

Latest release

Install led CLI

v4.7.0
amd64.deb · amd64View amd64 SHA-256
arm64.deb · arm64View arm64 SHA-256

Codex + Claude Code

Shared Agent SkillAdd one skill so your AI agent can publish, update, and inspect content with led CLI.
Agent Skill v4.7.0ledgerinfra-agent-skill-v4.7.0.tar.gz
View prompt
Install the LedgerInfra Skill from this URL in this project: https://github.com/system-witches/ledger-infra-releases/releases/download/led-v4.7.0/ledgerinfra-agent-skill-v4.7.0.tar.gz
Verify it with https://github.com/system-witches/ledger-infra-releases/releases/download/led-v4.7.0/ledgerinfra-agent-skill-v4.7.0.tar.gz.sha256 before extracting it. Use the Skill directory for this agent (Codex or Claude Code).
View Agent Skill SHA-256Add to Codex or Claude Code
Release notes

01 / PUBLISH

Start with a ready-made design proposal.

This fictional export-service proposal is one HTML file with no remote assets. You can use it without preparing company material. A new account already has a default Workspace.

Download the design proposalPreview the sample

Save or attach the downloaded file in your agent project, then paste this request:

View prompt
Use the LedgerInfra Skill. Publish the attached design-proposal.html as a new private Slide. Use my new account's default Workspace; if this CLI already has a selected Workspace, show it and ask before changing the destination. Do not create another Workspace.
If sign-in is required, start led login and let me approve the matching browser code.
Save the returned slide_key, version_key, latest URL, fixed version URL, and the actual Workspace key in ledgerinfra-quickstart-handoff.md in this project. Record the CLI profile name if one is used, but never save tokens or credentials in the handoff. Show me both URLs and ask me to review the proposal.

Check: Open the returned latest URL. You should see “Queue-backed export proposal” with 7-day retention. Keep the handoff file in the project; it contains references, not credentials.

02 / UPDATE

Change the proposal, keep the same review link.

In the same AI task, request one review change. Update the original Slide instead of creating another one.

View prompt
Use the LedgerInfra Skill and ledgerinfra-quickstart-handoff.md. Change the sample proposal's result retention from 7 days to 30 days, keeping export_job_id as the idempotency key and the limit of two concurrent workers. Update the sample's revision label to version 2. Save the changed HTML as design-proposal-v2.html.
Publish it as a new version of the SAME Slide in the SAME Workspace, preserving private visibility. Keep the first version's fixed URL and record the new version_key and fixed URL in the handoff. Confirm that the latest URL is unchanged. Show me the old and new fixed URLs so I can review the difference. Do not mark it approved for me.

Check: Reload the latest URL: it should show 30-day retention. Its address should match the first publish.

03 / REVIEW THE FIXED VERSION

Choose the version the next task should use.

Open both fixed URLs. The first should still show 7 days; the revised version should show 30 days. After reviewing the revised proposal, tell your agent which version you checked so it records that version key and fixed URL as your selected source in ledgerinfra-quickstart-handoff.md.

Check: You have opened the selected fixed version and checked its constraints. A published version is not an automatic approval.

How to check the URLs with the CLI
led show <SLIDE_KEY> --json
led versions <SLIDE_KEY> --json

Use the returned view_url for the latest version. Append /v/<version_key> to that URL, preserving its host, for a fixed version. Replace placeholders with your returned identifiers.

04 / START A NEW AI TASK

Use the reviewed proposal for the next change.

Open a new conversation in the same project with the Skill installed. Give it the handoff file and the request below. The saved CLI login must belong to the same person: this private sample is visible only to its owner.

View prompt
Use the LedgerInfra Skill. This is a new task. Read only the identifiers and URLs from ledgerinfra-quickstart-handoff.md, then retrieve the proposal from LedgerInfra in the same account and Workspace. Do not use the local sample HTML or the previous chat as the source.
Search for "Queue-backed export proposal" and use led fetch with the version_key I reviewed in the handoff to retrieve that exact HTML. Search indexing may be pending; if so, fetch the recorded fixed version directly and report that Search is pending. Use a new output filename rather than overwriting an existing file.
Using the retrieved proposal, draft a retry design for failed exports that avoids duplicate output. Explain how the proposal's retention period, idempotency key, and concurrency limit constrain your design. Cite the fixed version URL, distinguish source facts from your new recommendations, and show any missing information. Do not implement or publish the new design yet.

You are done when the new task uses the source.

Check that the retry design cites your selected fixed URL and uses the retrieved proposal’s 30-day retention, export_job_id idempotency key, and two-worker limit. It should explain how those constraints affect the design.

A successful search or fetch alone is not completion. If a constraint or citation is missing, ask the agent to read the fixed version again and correct the design before moving on.

Optional: share with a teammate after your first reuse

The first walkthrough stays private. For team review, first invite the reviewer to the owning Workspace, then deliberately change the Slide to workspace visibility. This changes access to earlier fixed versions too. Read-only agent grants do not include private Slides.

Read the invitation and access instructions

CONTINUE FROM WHERE YOU STOPPED

Find your current step.

Setup: the CLI or Skill is not found

Reopen the terminal after installing, then run led --version. Check that the Skill was installed in this project for your agent. Windows has no published installer; use a supported macOS or Ubuntu / Debian environment.

Check installation and Skill locations
Sign-in: email verification or browser approval is incomplete

For email/password signup, open the verification email, return to the account page and select Check verification. Use Resend email if needed. For an expired CLI login, run led login again and compare the new codes. After success, continue with your current step.

Open account access
Publish: it failed or the result is unclear

Ask your agent to check led whoami --json and led list --json in the intended Workspace. If the Slide already exists, recover its identifiers with led show and led versions before trying again. Keep successful publish URLs; do not create duplicate Slides to repair a later search failure.

Check file requirements and publish options
Review or reuse: the private source is inaccessible

Use the account that published the private sample. Confirm the Workspace and profile recorded in your handoff. A permission error requires correcting account or Workspace access; making the file public is not a recovery step.

Check Workspace and visibility rules
Reuse: Search is empty, pending, or unavailable

Publishing can succeed before indexing finishes. Keep the saved Slide and version keys. Ask the agent to check the query and Workspace, and fetch the reviewed fixed version directly while Search is unavailable:

led fetch <SLIDE_KEY> --version <VERSION_KEY> --output ./reviewed-proposal.html --workspace <WORKSPACE_KEY> --json

Use the identifiers from the handoff and a new output filename if one already exists. Resume the new-task prompt after the source is retrieved. Report missing source content instead of inventing it.