self-hosted-release-validation — Claude Code skill
self-hosted-release-validation is a Claude Code skill that validates OSS self-hosted releases via Docker Compose. Evaluates SQLite persistence, agent sources, and beta readiness.
Apache-2.0 ★ 104 updated 2026-11-17
npx skills add dodo-labs-ai/mem-forge --skill self-hosted-release-validation -a claude-codeWhat self-hosted-release-validation does
The skill validates self-hosted releases along the documented newcomer journey. It enforces an isolated Docker Compose project, loopback-only port bindings, a dedicated SQLite volume, UI and API startup, model configuration, source synchronization, and local MCP retrieval.
Execution relies on scripts/compose_smoke.sh. For model inference and embeddings, the script takes --ollama, an exported OPENROUTER_API_KEY, or explicit --generation-provider and --embedding-provider flags. Official release steps combine it with --quality alongside provider_smoke.py as detailed in docs/releasing-python-distribution.md.
Existing MemForge volumes or caches must remain untouched; tests run strictly on isolated data volumes. Persistence is verified by restarting the Compose stack against the existing volume and rechecking documents and MCP retrieval. The final review delivers a strict GO, CONDITIONAL GO, or NO-GO verdict.
What it helps with
- test clean Docker Compose startup before publishing a release
- verify SQLite volume persistence after restarting the service stack
- audit source connector matrices across servers and local daemons
- issue conservative GO or NO-GO decisions for public beta releases
Example prompts for the agent
- Run the self-hosted release validation using scripts/compose_smoke.sh in an isolated project
- Execute the Docker Compose smoke test with --quality and OPENROUTER_API_KEY
- Verify SQLite persistence across restarts and provide a beta readiness verdict
How to install self-hosted-release-validation
1Skills CLI (Vercel)
npx skills add dodo-labs-ai/mem-forge --skill self-hosted-release-validation -a claude-code2Manually: put the files in place
~/.claude/skills/self-hosted-release-validation/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/self-hosted-release-validation/
1Skills CLI (Vercel)
npx skills add dodo-labs-ai/mem-forge --skill self-hosted-release-validation -a codex2Manually: put the files in place
.agents/skills/self-hosted-release-validation/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add dodo-labs-ai/mem-forge --skill self-hosted-release-validation -a cursor2Manually: put the files in place
.cursor/skills/self-hosted-release-validation/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install dodo-labs-ai/mem-forge .agents/skills/self-hosted-release-validation2Manually: put the files in place
.github/skills/self-hosted-release-validation/SKILL.md1Skills CLI (Vercel)
npx skills add dodo-labs-ai/mem-forge --skill self-hosted-release-validation -a gemini-cli2Platform CLI
gemini skills install https://github.com/dodo-labs-ai/mem-forge --path .agents/skills/self-hosted-release-validationManually: put the files in place
.devin/skills/self-hosted-release-validation/SKILL.mdThe legacy .windsurf/skills/ path still works
Claude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about self-hosted-release-validation
How to install self-hosted-release-validation in Claude Code?
Run npx skills add dodo-labs-ai/mem-forge --skill self-hosted-release-validation -a claude-code. The definition is saved to ~/.claude/skills/self-hosted-release-validation/SKILL.md.
What does self-hosted-release-validation check?
It tests loopback Docker Compose startup, SQLite data persistence across restarts, real daemon jobs for local sources, and local MCP retrieval.
What is required to run self-hosted-release-validation?
You need Docker Compose, the repository script scripts/compose_smoke.sh, and either local Ollama or an exported OPENROUTER_API_KEY for embedding and generation models.
Source text
SKILL.md · Apache-2.0
--- name: self-hosted-release-validation description: "Validate an OSS self-hosted release through the newcomer Docker Compose path: isolated Compose project, SQLite persistence, source support matrix, local-agent sources and a public-beta GO/NO-GO. Use for compose smoke tests, release validation and beta-readiness reviews." --- # OSS Self-Hosted Release Validation - Validate self-hosted releases through the documented newcomer path: a clean Docker Compose project, loopback-only ports unless explicitly testing exposure, the default SQLite-backed data volume, Admin UI/API startup, first-run model configuration, a representative source sync, memory retrieval, and the packaged local MCP path. - `scripts/compose_smoke.sh` runs this newcomer path in an isolated Compose project (add `--ollama`, export `OPENROUTER_API_KEY`, or name each model kind's provider with `--generation-provider` and `--embedding-provider` for the model, sync and retrieval steps) and prints a table for the PR or release notes. A release runs it with `--quality`, beside `provider_smoke.py`, as the release guide, `docs/releasing-python-distribution.md`, states. - Use an isolated Compose project and dedicated volume for smoke tests. Never reuse or delete a developer's existing MemForge volume, source data, plugin cache, or external service state merely to obtain a clean result. - Prove SQLite persistence by restarting the API or Compose stack against the same isolated data volume and rechecking sources, documents, memories, keyword/vector health, and MCP retrieval. - Treat source support as an execution matrix, not a registry count. For each registered source type, state whether server collection, local-agent collection, credentials, browser sessions, VPN/internal reachability, or an external provider is required. Classify each type as verified, unsupported by design, blocked by a missing prerequisite, or failed. - Exercise local-agent sources through the real server-issued job and daemon path. A direct package injection may diagnose the service boundary, but it does not prove that the documented daemon workflow works. - For public-beta readiness, separately assess newcomer reproducibility, first-run UX, error messages, documentation accuracy, default data/privacy boundaries, and obvious exposure or secret-handling footguns. Give a conservative `GO`, `CONDITIONAL GO`, or `NO-GO`, with the smallest required fixes or disclosure language. - Do not present fixture-backed LLM/provider behavior as production-provider validation. Record the fixture contract, then distinguish it from real connector credentials and model-provider coverage.
© dodo-labs-ai · Apache-2.0 · dodo-labs-ai/mem-forge