refactor: simplify workspace committer — remove redundant parameters #17

New Issue

xiaoju · 2026-04-29T12:51:29Z

xiaoju commented

2026-04-29 12:51:29 +00:00

What

Simplify createWorkspaceCommitterRole in _shared/workspace-committer.ts. The agent can infer workflow name, commit scope, and branch naming from the thread history — no need to pass them as parameters.

Current signature

createWorkspaceCommitterRole(adapter, {
  extract,
  nerveRoot,
  workflowName,
  conventionalCommitScopeHint,
  branchCheckoutExample,
})

Target signature

createWorkspaceCommitterRole(adapter: AgentFn, extract: LlmExtractorConfig): Role<CommitterMeta>

Only adapter and extract remain. Everything else the agent figures out from the thread.

Changes

1. `_shared/workspace-committer.ts`

Remove BuildWorkspaceCommitterDeps type (or reduce to just extract)
Remove nerveRoot, workflowName, conventionalCommitScopeHint, branchCheckoutExample parameters
Simplify prompt: tell the agent to read the thread (nerve thread show <threadId>), infer the workflow context, and decide branch name / commit scope / message on its own
nerveRoot: get from start.meta.workdir or fall back to process.cwd(). The outer wrapper already sets start.meta.workdir — but actually, since we are simplifying, just let the agent use the working directory it is started in (the engine already sets workdir via the adapter). Remove the innerStart workdir override hack.
Keep the dry-run check and error handling wrapper
Keep the --author prohibition in the prompt

2. Prompt should be roughly

You are the committer agent. The coder finished with a passing build; your job is to branch, commit, and push.

1. Read the workflow thread: `nerve thread show <threadId>` — understand what was planned, coded, and reviewed.
2. Run `git status`. If nothing to commit, set committed=false.
3. Create a feature branch: infer a good `fix/<slug>` or `feat/<slug>` name from the thread context.
4. `git add -A`
5. Write a conventional commit message based on the thread context.
6. `git commit -m "<message>"` — do NOT pass `--author`, use repo git config.
7. `git push -u origin <branch>`

3. Update all callers

workflows/develop-sense/build.ts — change createWorkspaceCommitterRole(a("committer"), { extract, nerveRoot, ... }) to createWorkspaceCommitterRole(a("committer"), extract)
workflows/develop-workflow/build.ts — same
workflows/develop-sense/roles/committer.ts — update re-export if type changed
workflows/develop-workflow/roles/committer.ts — same
workflows/solve-issue/roles/committer/ — if it uses the shared function, update too. If it has its own committer, apply same simplification.

4. Verify

pnpm build at repo root must pass

## What Simplify `createWorkspaceCommitterRole` in `_shared/workspace-committer.ts`. The agent can infer workflow name, commit scope, and branch naming from the thread history — no need to pass them as parameters. ## Current signature ```ts createWorkspaceCommitterRole(adapter, { extract, nerveRoot, workflowName, conventionalCommitScopeHint, branchCheckoutExample, }) ``` ## Target signature ```ts createWorkspaceCommitterRole(adapter: AgentFn, extract: LlmExtractorConfig): Role<CommitterMeta> ``` Only `adapter` and `extract` remain. Everything else the agent figures out from the thread. ## Changes ### 1. `_shared/workspace-committer.ts` - Remove `BuildWorkspaceCommitterDeps` type (or reduce to just `extract`) - Remove `nerveRoot`, `workflowName`, `conventionalCommitScopeHint`, `branchCheckoutExample` parameters - Simplify prompt: tell the agent to read the thread (`nerve thread show <threadId>`), infer the workflow context, and decide branch name / commit scope / message on its own - `nerveRoot`: get from `start.meta.workdir` or fall back to `process.cwd()`. The outer wrapper already sets `start.meta.workdir` — but actually, since we are simplifying, just let the agent use the working directory it is started in (the engine already sets workdir via the adapter). Remove the `innerStart` workdir override hack. - Keep the dry-run check and error handling wrapper - Keep the `--author` prohibition in the prompt ### 2. Prompt should be roughly ``` You are the committer agent. The coder finished with a passing build; your job is to branch, commit, and push. 1. Read the workflow thread: `nerve thread show <threadId>` — understand what was planned, coded, and reviewed. 2. Run `git status`. If nothing to commit, set committed=false. 3. Create a feature branch: infer a good `fix/<slug>` or `feat/<slug>` name from the thread context. 4. `git add -A` 5. Write a conventional commit message based on the thread context. 6. `git commit -m "<message>"` — do NOT pass `--author`, use repo git config. 7. `git push -u origin <branch>` ``` ### 3. Update all callers - `workflows/develop-sense/build.ts` — change `createWorkspaceCommitterRole(a("committer"), { extract, nerveRoot, ... })` to `createWorkspaceCommitterRole(a("committer"), extract)` - `workflows/develop-workflow/build.ts` — same - `workflows/develop-sense/roles/committer.ts` — update re-export if type changed - `workflows/develop-workflow/roles/committer.ts` — same - `workflows/solve-issue/roles/committer/` — if it uses the shared function, update too. If it has its own committer, apply same simplification. ### 4. Verify - `pnpm build` at repo root must pass

xiaoju referenced this issue from a commit

2026-04-29 12:53:00 +00:00

refactor: simplify workspace committer — agent infers context from thread

xiaoju referenced a pull request that will close this issue

2026-04-29 12:53:07 +00:00

refactor: simplify workspace committer — infer from thread #18

xiaoju closed this issue

2026-04-29 12:53:35 +00:00

Sign in to join this conversation.