chore(todo): drop items completed in this session

Removes the three feature asks shipped this session:
- batch_commit push: "after" option
- git_merge tool
- git_cherry_pick tool

Also reframes the batch_commit stage-scoping item: the path-level
MVP (per-entry `files: string[]`) already exists, so only the
hunk-level stretch goal remains open. Other asks (worktree tools,
git_reset_soft, git_diff text, git_show, batch_commit dry_run,
stash/tag/fetch, atomic-stage doc note, JSON schema publishing)
are untouched — they remain genuine gaps in the MCP surface.

Author: AlbinoGeek

TODO.md

@@ -0,0 +1,96 @@
+# TODO / Feature requests — rethunk-git MCP
+
+Feature asks driven by real pain points from agent sessions. Each item lists the motivating scenario and the expected tool shape.
+
+## High value — reduce Bash fallback
+
+### `batch_commit` — `stage` argument for path/hunk scoping
+
+**Pain:** `batch_commit` stages whole files atomically. Subagents asked to produce N commits from a single file edit cannot split post-hoc (sandbox blocks `git reset --soft`; `git rebase -i` is interactive).
+
+**Ask:** Allow scoped staging:
+
+```ts
+batch_commit({
+  stage: { paths: ["Makefile"], hunks: [{ file: "Makefile", lines: "120-180" }] },
+  message: "feat(make): add sbom target"
+})
+```
+
+Path-level `files: string[]` already exists per-entry on `batch_commit`. Remaining gap is hunk-level staging via `git add -p` patch interface.
+
+### `git_worktree_add` / `git_worktree_remove` / `git_worktree_list`
+
+**Pain:** Parallel subagent batches use worktrees heavily. Today: Bash `git worktree add/remove/list`. Should be MCP-native.
+
+**Ask:**
+
+```ts
+git_worktree_add({ path: ".claude/worktrees/agent-abc", branch: "worktree-agent-abc", baseRef: "main" })
+git_worktree_remove({ path: "...", force?: false })
+git_worktree_list() // → [{ path, branch, commit, locked }]
+```
+
+### `git_reset_soft` MCP tool
+
+**Pain:** Sandbox blocks Bash `git reset --soft HEAD~N`, but this is the only way to split atomic commits. Subagents hit this repeatedly.
+
+**Ask:** Narrow, safe MCP tool:
+
+```ts
+git_reset_soft({ ref: "HEAD~3" })
+// Reset branch pointer by N commits while keeping staged index.
+// Refuses if working tree is dirty (reserve for workflows where all changes are already committed).
+```
+
+## Medium value — ergonomics
+
+### `git_diff` — scoped
+
+**Pain:** `git_diff_summary` is aggregate; sometimes need the actual diff text for a specific file/range. Currently Bash fallback.
+
+**Ask:**
+
+```ts
+git_diff({ paths?: string[], base?: "HEAD~1", target?: "HEAD", unified?: 3 })
+```
+
+### `git_show` MCP tool
+
+**Pain:** Inspecting a subagent branch's commits uses Bash `git show <sha>`. Should be covered.
+
+**Ask:**
+
+```ts
+git_show({ ref: "sha", stat?: boolean, paths?: string[] })
+```
+
+### `batch_commit` — `dry_run: true`
+
+**Pain:** Agent wants to verify staging is what it thinks before committing. Today: commit, inspect, reset (blocked), retry.
+
+**Ask:** `dry_run: true` reports what would be committed (files, message, diff summary) without writing.
+
+## Low value — nice to have
+
+### `git_stash_apply` / `git_stash_list` MCP tools
+
+Currently Bash. Stash flows come up occasionally in agent sessions when a conflict needs to be set aside.
+
+### `git_tag` MCP tool
+
+For release workflows: `git_tag({ name: "v0.6.0", message: "Release 0.6.0", signed?: false })`.
+
+### `git_fetch` MCP tool
+
+With structured output: `{ updated: [{ ref, oldSha, newSha }], newRefs: [...] }`.
+
+## Non-tool asks
+
+### Document `batch_commit` atomic-stage semantics
+
+In `docs/` README: state clearly that `batch_commit` stages listed files atomically before creating the commit, so N back-to-back calls to `batch_commit` on the same file cannot produce N commits with distinct content. Agents routinely expect per-call incremental semantics.
+
+### Publish JSON schemas for all MCP tool args
+
+Helps agents validate calls locally before invoking. Also eases codegen for skill builders.