finishing-a-development-branch
Use after execution is complete or intentionally stopped — decide whether to merge back, publish a PR branch, keep the execution checkout, or discard it, and…
Skill by yimwoo
Install
Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.
npx agentmarket install skill/yimwoo-finishing-a-development-branch
Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.
Get the weekly diff
The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.
About
Reviews
Write a reviewNo reviews yet. Installed it? Reviews come from verified installers only. How reviews work
Permissions
What this skill can do and what it asks for, from its SKILL.md. Checked Oct 25, 2026.
- Tools
- No tools found
- Secrets
- None asked for
- Install scripts
- None found
Network hosts it talks to 1
codeload.github.com· found in install command
Files it touches
~/.claude/skills/finishing-a-development-branch
What's inside
- Skill name
finishing-a-development-branch- Description
- Use after execution is complete or intentionally stopped — decide whether to merge back, publish a PR branch, keep the execution checkout, or discard it, and record that disposition in HOTL state.
- License
- MIT
- Instructions
- 101 lines in SKILL.md · view on GitHub
Files (1)
- SKILL.md
SKILL.md (first lines)
# Finishing A Development Branch Use this after `loop-execution`, `executing-plans`, or `subagent-execution` when the run already has a `run_id` and you need to decide what happens to the execution branch/worktree next. ## Core Idea HOTL distinguishes between: - **Authoring checkout** — where the workflow was written (`source_branch`, `source_head`) - **Execution checkout** — where the workflow ran (`branch`, `execution_root`, `worktree_path`) Finishing is the stage that closes that loop intentionally. Do not silently merge, delete, or abandon the execution worktree. ## Step 1: Read Execution Provenance Run: ```bash hotl-rt summary <run-id> --json ``` Surface these fields before presenting options: - `status` - `source_branch` - `branch` - `execution_root` - `worktree_path` - `finish.disposition` (if already recorded) - controller owner/status and receipt reasons If `finish.disposition` is already set, stop and tell the user the run is already finished. A successful execution must be `ready_to_finish` before merge or publish. `ready_to_finish` means execution evidence is complete but the run is not yet `completed`. Keep the existing controller lease active and retain `HOTL_OWNER_TOKEN`; if a different controller is finishing, use explicit owner handoff/takeover rules from `resuming` first. Run `owner heartbeat` before a long finish operation. ## Step 2: Present The 4 Finish Options Use these options: 1. **Merge back locally** - Default target: `source_branch` when present and different from the execution branch - Fallback target: `main` or `master` 2. **Publish branch / create PR** - Push the execution branch to a remote - If `gh` is available and the user wants a PR, create it against the chosen target branch 3. **Keep as-is** - Preserve the execution branch/worktree for later review or follow-up changes 4. **Discard** - Remove the execution branch/worktree after explicit confirmation Explain the target branch explicitly when merge or PR target is inferred from `source_branch`. ## Step 3: Authorize External Effects For publish/PR, record one bounded `external_write` with the exact target `publish <branch> to <remote>` (append ` and create PR against <target-branch>` when `--create-pr` is used), pass `--idempotency-key <stable-key>`, obtain human `action decide ... approved --mode human`, and pass the returned action id and key to the helper. If a merge is classified as `production_change`, use the exact target `merge <branch> into <target-branch>`. The helper verifies kind, target, key, approval, and not-started status; it then performs `action begin` immediately before the effect and `action complete` with observed evidence before recording finish. Approval is not effect evidence. If the helper stops after a push, PR attempt, or merge may have started, leave the action unresolved, inspect the target, and use `action reconcile`. Never rerun the helper while the effect is `in_progress` or `uncertain`. ## Step 4: Execute Via The Helper Use the repo-owned helper: ```bash scripts/hotl-finish-execution.sh --run-id <run-id> --mode <keep|merge|publish|discard> ... ``` Mappings: - Keep → `--mode keep` - Merge locally → `--mode merge [--target-branch <branch>]` - Publish / PR → `--mode publish [--remote origin] [--target-branch <branch>] [--create-pr]` - Discard → `--mode discard --confirm discard` For an approved merge/publish effect, also pass `--effect-action-id <id> --idempotency-key <key>`. The helper records intent with `action begin`, executes the bounded effect, persists `action complete` evidence, then records the finish disposition with `hotl-rt finish ...`. It enforces safety checks around branch switching, merging, cleanup, and destructive discard. `HOTL_OWNER_TOKEN` must remain in the helper environment. ## Step 5: Report The Outcome After the helper succeeds, show: - Finish disposition - Target branch / PR URL if applicable - Whether the execution worktree was kept or removed - Where the durable report lives Then run `owner release` and derive a fresh receipt. Claim completion only when the run is `completed` and `sufficiency.sufficient` is true. A failed helper must leave finish unset; use `action reconcile` before another attempt when an external effect may have occurred. Important behavior: - Merge and discard preserve `.hotl/state/<run-id>.json` and `.hotl/reports/<run-id>.md` back into the repo checkout before removing an isolated execution worktree - Publish keeps the execution worktree by default so review follow-ups can continue on the same branch - Shared-checkout same-branch runs are intentionally conservative; discard and merge may require manual handling if automation would be ambiguous or destructive ## Guardrails - Do not auto-merge after execution just because tests passed - Do not auto-delete work without explicit user confirmation - Do not create a PR unless the user chose the publish/PR path - If merge conflicts occur, stop and leave the execution branch/worktree intact - If a chosen PR operation cannot run because `gh` is unavailable or fails after push, do not silently downgrade the approved target. Leave finish unset, reconcile the observed push, and ask whether the user wants a new bounded push-only action.
From yimwoo/hotl-plugin, MIT.
More from yimwoo/hotl-plugin
- 0.559score
receiving-code-review
yimwoo
Use when review findings arrive — verify each claim against the codebase and HOTL contracts before making changes.
SkillCFreshness: C★ 27Claude Code +9 - 0.559score
skill-authoring
yimwoo
Use when creating, editing, or reviewing HOTL skills, agents, command prompts, or other behavior-shaping instruction files.
SkillCFreshness: C★ 27Claude Code +9 - 0.558score
dispatch-agents
yimwoo
Use when you have 2+ independent tasks with no shared state — dispatches parallel subagents for each task.
SkillCFreshness: C★ 27Claude Code +9 - 0.556score
subagent-execution
yimwoo
Delegated step runner over the HOTL execution state machine — delegates eligible steps to fresh subagents while the controller keeps governance, verification…
SkillCFreshness: C★ 27Claude Code +9 - 0.556score
verification-before-completion
yimwoo
Use before claiming any work is complete, fixed, or passing — runs verification commands and confirms output before making success claims.
SkillCFreshness: C★ 27Claude Code +9 - 0.556score
pr-reviewing
yimwoo
Review a PR across multiple dimensions — description, code changes, code scan, unit tests — using parallel subagents.
SkillCFreshness: C★ 27Claude Code +9 - 0.556score
requesting-code-review
yimwoo
Use at executor review checkpoints to dispatch the code-reviewer agent with structured context — git range, workflow contracts, and verification evidence.
SkillCFreshness: C★ 27Claude Code +9 - 0.555score
loop-execution
yimwoo
Use when executing an existing HOTL workflow file — reads steps, loops until success criteria met, auto-approves low-risk gates, pauses at high-risk gates.
SkillCFreshness: C★ 27Claude Code +9 - 0.554score
executing-plans
yimwoo
Use when executing an implementation plan linearly with explicit human checkpoints between batches of tasks.
SkillCFreshness: C★ 27Claude Code +9 - 0.554score
governed-execution
yimwoo
Use when executing an accepted HOTL workflow through the best available host or fallback driver.
SkillCFreshness: C★ 27Claude Code +9 - 0.553score
tdd
yimwoo
Use before writing any implementation code — enforces RED-GREEN-REFACTOR cycle.
SkillCFreshness: C★ 27Claude Code +9 - 0.553score
code-review
yimwoo
Use after completing implementation steps and before merging — reviews against plan and HOTL contracts.
SkillCFreshness: C★ 27Claude Code +9
Signals
- Stars
- 27
- Forks
- 3
- Open issues
- 1
- Last commit (this folder)
- Jul 6, 2026 (3 months ago)
- Last release
- Jul 2, 2026 (4 months ago)
From GitHub (fetched Oct 24, 2026).
Details
- License
- MIT
- Repository
- github.com/yimwoo/hotl-plugin
- First seen
- Oct 24, 2026
- Last checked
- Oct 24, 2026
- Source
- GitHubgithub-graphql
More from yimwoo
- 0.559score
receiving-code-review
yimwoo
Use when review findings arrive — verify each claim against the codebase and HOTL contracts before making changes.
SkillCFreshness: C★ 27Claude Code +9 - 0.559score
skill-authoring
yimwoo
Use when creating, editing, or reviewing HOTL skills, agents, command prompts, or other behavior-shaping instruction files.
SkillCFreshness: C★ 27Claude Code +9 - 0.558score
dispatch-agents
yimwoo
Use when you have 2+ independent tasks with no shared state — dispatches parallel subagents for each task.
SkillCFreshness: C★ 27Claude Code +9 - 0.556score
pr-reviewing
yimwoo
Review a PR across multiple dimensions — description, code changes, code scan, unit tests — using parallel subagents.
SkillCFreshness: C★ 27Claude Code +9 - 0.556score
requesting-code-review
yimwoo
Use at executor review checkpoints to dispatch the code-reviewer agent with structured context — git range, workflow contracts, and verification evidence.
SkillCFreshness: C★ 27Claude Code +9 - 0.556score
subagent-execution
yimwoo
Delegated step runner over the HOTL execution state machine — delegates eligible steps to fresh subagents while the controller keeps governance, verification…
SkillCFreshness: C★ 27Claude Code +9
Cite this page
finishing-a-development-branch skill. Agent Mart. https://agentmart.dev/skills/yimwoo-finishing-a-development-branch (last updated 2026-10-24)