On The Record — Git skill for Claude Code
Everything on the record: a requirement is an issue, a decision is an approval comment, work is a PR, rationale is that PR's record.
How to install On The Record
This entry records only its repository, not the path inside it, so there is no
exact command to give. Open tokenmaxxxer/on-the-record and copy the folder into
~/.claude/skills/, or the file into ~/.claude/agents/.
What On The Record does
Everything on the record: a requirement is an issue, a decision is an approval comment, work is a PR, rationale is that PR's record. The orchestrator relays conversation to GitHub and spawns sandboxed role sessions, each with only its own rulebook.
Alternatives in Git
- Maintainers — Jarrod Watts (https://github.com/jarrodwatts) If you are interested in becoming a maintainer, open an issue to 10.9k ★
- Fresh Start — The original nirholas/claude-code before DMCA and take down 6.2k ★
- Prepare Commit — Comment out inappropriate logging before commit 1.6k ★
README
tokenmaxxxer / on-the-record
*[한국어](README.ko.md)*
Quickstart
gh auth login
In your conversational session:
/plugin marketplace add tokenmaxxxer/on-the-record
/plugin install on-the-record@tokenmaxxxer
/on-the-record:run
No separate clone or secret setup is required for on-the-record, rulebooks, or tokenmaxxxer-core. **skill-repository is the exception**: all roles resolve their guidance from it, and it needs a manual clone with `MUSTER_SKILL_REPO` pointed at the checkout's `skills/` directory (or a sibling clone for zero-config) — see the skill-repository step in [`docs/handbooks/setup.md`](docs/handbooks/setup.md) (see "Getting started").
Full requirements and optional setup (agent account, model pinning, project init) are documented in [`docs/handbooks/setup.md`](docs/handbooks/setup.md).
Interaction flow
This section describes the mechanics after install: issues, PRs, branches, records, gates.
The user-facing loop
The user states a need in conversation. The orchestrator drafts an issue and relays it to GitHub under the user's account. The user confirms in conversation. From that point, user input and AI activity are each recorded in one fixed location: a requirement is an issue, a decision is an approval comment, work is a PR on an `issue-/` branch, and the rationale is that PR's record document. A role ships phase 1 (proposal) first; phase 2 (delivery) opens only after the user approves. The user does not write to GitHub directly. Every decision is made in conversation with the orchestrator, which relays it to GitHub under the user's account: feedback as a comment, approval as an `APPROVE issue-/` comment, acceptance as a merge, rejection as a close. The orchestrator also watches for each PR the role opens and reports its contents back to the user in conversation before the user decides.
sequenceDiagram
participant U as User
participant O as Orchestrator
participant R as Role sessi
Related Skills
Shipped
The record: what reached the integration branch, when, and which decision or request each change was for. Plus
Reject
Reject a PR — posts feedback comment, removes approval labels, routes back to Builder
V Triage
Classify one change request before any work starts — resolve, score, and write plus COMMIT the pre-eval triage
Gw Commit Context
Load commit context for Claude to understand implementation details and rationale.
Claude CLI Telegram
Turn your Mac/Linux into a remote Claude Code terminal via Telegram. Spawns real CLI — sessions, streaming, ap
Merge Conflict Walkthrough
Walks through each conflicted file in a merge or rebase - lists files, reads markers, proposes a resolution wi
Related Agents
Wave Lead
Leads one wave of ai-training issues for the /wave dispatcher. Spawns one builder per issue and one reviewer p
Alfresco Architect Agent
Given a business requirement, proposes full Alfresco extension architecture including content types, behaviour
Human Comment Reader
Comment-reading agent — reads issue/PR threads like a person: maintainer tone, repo culture, implicit rules —