Generate release notes from git commit history

Self-hosted autonomous agent that turns your commit history into clean, grouped, human-readable release notes — durably, so a release-notes job is never lost.

Overview

This pack provides an autonomous agent that turns a list of commit messages into a structured changelog (Keep a Changelog style): commits are classified according to the Conventional Commits convention (feat, fix, perf, refactor, docs…), grouped by section, breaking changes are highlighted, and the result is rendered as Markdown ready to paste into a CHANGELOG.md or a GitHub/GitLab release.

The core of the product is a durable SQLite job queue. Each release generation is a task: a success is archived, a transient failure is rescheduled with backoff, a terminal failure goes to the dead-letter queue for inspection or replay. An idempotency key prevents generating the notes for the same version twice.

This pack targets operators who ship regular releases (libs, SaaS, CLIs, agent packs) and want clean release notes without manual writing. It does not replace a full CI/CD tool and does not execute any arbitrary git command supplied in the payload: it consumes a list of commits that you pass to it.

The pack is designed to be self-hosted (BYO-LLM): no API key is included. The LLM adapter detects a local CLI (Claude Code / Kimi Code / Codex) and uses it only to rewrite the wording of entries into clear prose. The deterministic parser remains the source of truth: the LLM can neither invent nor remove a change.

Architecture

┌─────────────────────────────────────────────────────────────────────┐
│                    Commit list (payload / log)                       │
└──────────────────────────────────┬──────────────────────────────────┘
                                   ▼
┌─────────────────────────────────────────────────────────────────────┐
│  IDLE → COLLECTING → CATEGORIZING → DRAFTING → REVIEWING → PUBLISHING │
│                           OODA State Machine                         │
└──────────────────────────────────┬──────────────────────────────────┘
                                   ▼
┌─────────────────────────────────────────────────────────────────────┐
│  Conventional-commit parser (deterministic)  →  BYO-LLM Editor       │
│  load_prompt("Editor")  →  wording rewrite only                      │
└──────────────────────────────────┬──────────────────────────────────┘
                                   ▼
┌─────────────────────────────────────────────────────────────────────┐
│  SQLite queue  →  retry / dead-letter  →  rendered Markdown changelog │
└─────────────────────────────────────────────────────────────────────┘

Pack contents

Quick installation

cd /opt/agent-changelog-v1
sudo bash install.sh

The installer checks for Python 3.10+, creates a venv, installs the dependencies, detects a local LLM CLI and configures a systemd unit.

Immediate test

# Offline test suite (deterministic, no API key)
python3 test_agent.py

# Generate a changelog from commits, without an LLM
python3 agent.py enqueue --type render_changelog \
  --payload '{"version":"1.2.0","commits":["feat(api): add CSV export","fix: crash on empty input","feat!: drop py38"]}'
python3 agent.py run --once --deterministic

Job types

• categorize_commits: classifies a list of commits into ordered sections.
• draft_release: produces a structured release object (version + date + sections).
• render_changelog: renders the final Markdown changelog.

python3 agent.py enqueue --type render_changelog \
  --payload '{"version":"2.0.0","date":"2026-06-23","commits":["feat: ...","fix: ..."]}'
python3 agent.py run --once
python3 agent.py status
python3 agent.py replay --task t-ab12cd34

Deterministic mode covers all parsing and rendering without an LLM. The BYO-LLM mode is limited to improving the wording of entries: its output is a JSON object {"markdown": "..."} and any error falls back to the deterministic rendering.

Determinism vs LLM

Customization

1. Copy scripts/config.yaml to config.local.yaml.
2. Adapt templates/changelog_config.yaml (sections, order, type mapping).
3. Plug in your commit source — see CUSTOMIZE.md.

Costs and limits

This pack favors deterministic logic: everything works without an LLM. The LLM is only used to polish the text. See OPERATING_COST_ESTIMATE.md for the operating cost estimate.

Use a local model (Ollama)

The optional LLM step can also run on your own machine through Ollama, so no text leaves your server. No account and no API key.

ollama pull <model>               # any chat model you like
export PACK_LLM=ollama            # prefer Ollama even if a coding-agent CLI is installed
export PACK_OLLAMA_MODEL=<model>  # optional: default is the first chat model the server lists
python3 -c "import llm_adapter; print(llm_adapter.detect())"   # prints: ollama

• PACK_OLLAMA_HOST points to another server (default http://127.0.0.1:11434); OLLAMA_HOST is also read when PACK_LLM=ollama.
• PACK_OLLAMA_NUM_CTX sets the context size if your reports or prompts are long.
• Without PACK_LLM, a local Ollama server is used only when no Claude Code, Kimi Code or Codex CLI is found.
• If the server is down or the model is missing, the pack keeps working in its deterministic mode.

Learning memory (local and private)

Every LLM call this pack makes is journaled in ./.pack_memory/ next to the code, so the pack can improve for you. Nothing is ever sent anywhere.

• What is kept: the task name, the masked input, the masked answer, status and latency. Keys, tokens, e-mails, phone numbers, IP addresses, card and IBAN numbers and personal paths are masked before writing; the system prompt is never stored.
• Teach it: python3 pack_memory.py feedback last ok confirms a result, python3 pack_memory.py feedback last bad --fix "the right answer" corrects it. Confirmed and corrected runs are reused as validated examples when a similar input comes in. Unrated runs are never injected into a prompt.
• Find similar cases: python3 pack_memory.py similar "some text" and python3 pack_memory.py clusters (groups of near-identical failures) work in plain Python, no model needed. Optional meaning-based matching uses an embedding model you already run locally: python3 pack_memory.py enable-semantic (Ollama on this machine only).
• Share, only if you want: python3 pack_memory.py export writes an anonymised summary (counts, statuses, size buckets, no text) for you to review.
• Control: PACK_MEMORY=0 turns it off, PACK_MEMORY_STORE_TEXT=0 keeps fingerprints only, python3 pack_memory.py purge --yes wipes it.

License

Personal use per purchase. Resale prohibited.