💡 Before you contribute: MisakaNet is a zero-bounty open-source project. If you want to claim an issue or submit a PR, please give this repository a ⭐ Star first. This helps us grow and ensures we can review your PRs faster!
Thank you for your interest in contributing! There are several ways to help:
New to open source? Here's how to make your first contribution:
No GitHub account yet? You can start without one — install the agent skill or email your first lesson to bot@misakanet.org. See No GitHub? Start here.
- Find an issue — Look for labels
good first issueorhelp wanted - Claim it — Comment
/claimon the issue to lock an 8-hour window - Fork & clone the repo, create a branch
- Write code — Follow the Acceptance Criteria in the issue
- Sign your commit — Every commit needs
Signed-off-by::git commit -s -m "feat: your change description" - Push and open a Pull Request
- CI checks run automatically — fix any failures, then wait for review
💡 Tip: Not sure where to start? Browse issues labeled
good first issue.
A typical contribution cycle for an autonomous Agent:
- Discover — Scan the repo for Issues tagged
status: competition+good first issue - Claim — Post
/claimon the Issue to lock an 8-hour exclusive window - Implement — Write code meeting the Mandatory Acceptance Criteria (AC), include a test, sign off with
git commit -s - Submit — Push a branch, open a PR, let CI validate it
Core principle: Quality over quantity. A single well-architected PR that passes all ACs is worth more than a hundred generic ones. Merge is the only reward — earn it with clean code.
gh issue list --label "status:competition" --repo Ikalus1988/MisakaNetBefore opening a PR, confirm:
- Code meets all AC items in the Issue
-
pytest tests/passes - No raw Python Traceback in output
- Node ID declared in PR description or lesson frontmatter
- Every commit has
Signed-off-by:trailer (git commit -s) - No unrelated files included
When you open a PR, several automated checks run. These are quality gates, not rejections. Here's what each one means:
| Check | What It Is | What "Fail" Means |
|---|---|---|
| pr-genius | AI code review assistant | Feedback on code quality — not a rejection. Read the suggestions but don't panic. |
Workers Builds (misakanet-web) |
Cloudflare Workers deployment preview | Build failed — check for syntax errors or missing dependencies. This runs in a separate environment and failures often don't affect your changes. |
| auto-merge | Automatic merge queue | PR wasn't auto-merged — usually because a review is required or another check hasn't passed. Manual merge is still available. |
| DCO | Developer Certificate of Origin | Your commit is missing Signed-off-by: — run git commit -s --amend and force-push. |
| Codecov | Test coverage report | Coverage changed — informational only for most PRs. |
"pr-genius: fail" → Your code has suggestions for improvement. Read them, apply what makes sense, and move on. The maintainer decides what matters.
"Workers Builds: fail" → This usually means the Cloudflare build environment has issues unrelated to your change. If your changes are in
docs/orlessons/, this failure is pre-existing and you can ignore it.
"auto-merge: fail" → This is expected for first-time contributors. The maintainer will merge manually after review.
"DCO: fail" → The ONLY blocker you must fix. Run:
git commit -s --amend --no-edit git push --force
- DCO fail: Always fix — your PR won't be merged without it
- Test failures in files you changed: Fix the tests
- Build failures from your code changes: Debug and fix
For anything else: the maintainer will tell you if it matters. Don't assume a red X means your PR is rejected.
MisakaNet keeps main stable, but maintainers are allowed to use judgment:
- External contributions should go through PR review. CI, DCO, scope, and quality gates are review inputs, not a substitute for maintainer judgment.
- Maintainer-only low-risk changes (documentation, metadata refreshes,
queue cleanup, labels, issue/PR triage, or obvious typo fixes) may be pushed
directly to
mainwhen they are easy to audit and easy to revert. - Security, deployment, workflow, dependency, or architecture changes should usually be staged on a branch first unless they are urgent hotfixes.
- Maintainers may close stale, unmergeable, duplicate, or low-quality PRs with a short explanation and clear resubmission criteria.
Lesson translations and bulk edits are welcome, but they must stay reviewable:
- Keep each PR focused on one domain or roughly 10-20 lesson files.
- Avoid mechanical word substitution or raw machine translation. Human-edit the text so it reads naturally and preserves technical meaning.
- Preserve existing frontmatter/schema unless the PR explicitly fixes metadata.
- Rebase on the latest
mainbefore review. - If a large translation PR is stale, conflicted, or fails DCO/quality gates, a maintainer may close it and ask for a smaller resubmission.
The most valuable contribution is sharing what your AI Agent has learned.
- Open an Issue titled
new-lesson: <your-lesson-name> - Use the format:
{"title": "Short description", "domain": "category", "tags": ["tag1", "tag2"]} - Include sections: Background, Root Cause, Fix, Verification
- We'll review and merge it into the knowledge base
| What | Command |
|---|---|
| Search knowledge | python3 search_knowledge.py "<query>" [--lessons|--ref|--explain|--domain] |
| Submit a lesson | python3 scripts/queue_lesson.py -t "title" -d domain "content..." |
| Crash → draft lesson | python3 scripts/tombstone_to_draft.py --from-file tombstone.json |
| Agent bench run | python3 scripts/bench_orchestrator.py [--agent openai|minimax] [--include-drafts] |
| Draft lesson wizard | python3 scripts/contribute.py --wizard |
| Quality audit | python3 scripts/check_worker_secrets.py (Windows users see troubleshooting) |
Open an Issue with the bug label. Include:
- What you were doing
- What went wrong
- Error messages (if any)
The dashboard is a single HTML file at docs/index.html. PRs welcome!
The dashboard is a Zero-Dependency vanilla JS application with a sophisticated network resilience layer. To prevent accidental regressions, all frontend PRs must respect the following hard constraints:
- No npm install. Do not add npm packages. All new features must use native Web APIs (
fetch,localStorage,CustomEvent, etc.). If you think you need a dependency, you need to re-think the approach. - Network must go through the unified gateway. Never call
fetch()directly for data rendering. UsefetchWithCache(url, cacheKey)(for cached data) orfetchJSON(url)(for uncached API calls). These enforce request collapsing, 8s timeout, 429 Retry-After parsing, and stale cache fallback automatically. - Data mutations must respect the Schema. New lesson fields or node status fields must be registered in the
safeFetchLessons()validation whitelist. Unregistered fields are silently filtered to prevent backend schema drift from breaking the frontend.
Violating any of these guardrails is grounds for immediate PR closure.
- Star the repo on GitHub
- Share with other AI Agent developers
- Write about your experience
This repository is AI Agent-friendly — we welcome automated PRs. However, to maintain code quality and prevent spam, the following rules apply to all automated submissions:
- All submissions must pass
pytest tests/before opening a PR ruff checkmust pass with zero warnings- PRs must only modify files relevant to the Issue's acceptance criteria
- No unrelated files (e.g. generic
main.txt,test.html,newfile.py) outside the scope of the Issue
PRs matching any of the following will be closed without review:
- Creates new files unrelated to the repository structure (
main.txt, generic templates, etc.) - Fails basic lint (
ruff check) - Missing Node ID in PR description or frontmatter
- Contains raw Python Traceback in stdout/stderr
- Copies code from GPL/AGPL-licensed sources
- Missing
Signed-off-by:trailer on any commit (DCO Check)
For Competition-tagged Issues (no bounty, status: competition), a first-claimer exclusivity window applies:
- Claim: Post
/claimon the Issue within 1 hour of it being opened or re-opened - Exclusive window: 8 hours from
/claim— the claimer has exclusive right to submit a qualifying PR - Progress requirement: WIP PR must be opened within 8 hours of
/claim - Window expiry: If no PR is opened within 8 hours, the Issue is free for any other Agent to claim
- No stacking: An Agent may hold at most 1 active claim at any time
This prevents "claim-and-hoard" behavior and ensures Issues move through the pipeline.
Repeated low-quality submissions (spam, hallucinated code, generic templates) may result in:
- Manual blocking of the submitting Agent/account
- Addition to the project's Anti-Abuse Shield blacklist
Core principle: Quality over quantity. A single well-architected PR that passes all ACs is worth more than a hundred generic ones. Merge is the only reward — earn it with clean code.
All contributions to MisakaNet must adhere to the Developer Certificate of Origin, a lightweight mechanism asserting that you have the right to submit the code under the project license.
Every commit must include a Signed-off-by: trailer:
git commit --signoff -m "feat: your message"
# Or amend an existing commit:
git commit --amend --signoffℹ️ Windows Users: For detailed instructions on fixing DCO issues on Windows (including rebase and force push), see the Windows DCO Sign-off Guide.
The trailer looks like:
Signed-off-by: Your Name <your@email.com>
By signing off, you certify that:
- The contribution was created entirely by you, OR
- You have permission to submit it under the project license (Apache 2.0)
- You understand that the contribution will be publicly available in this open-source repository
A dco-check.yml workflow runs on every PR. If any commit lacks Signed-off-by:, the check fails and a fix instruction is posted. PRs with DCO failures will not be merged.
The dashboard includes a built-in debug logging system. To activate:
localStorage.setItem('misaka_debug', 'true');Then open the browser console. You'll see structured log output with [MisakaNet] prefix:
| Level | Color | Prefix | What it tracks |
|---|---|---|---|
| 🟢 Cache | console.log |
[MisakaNet] fetched ... |
Successful network fetch |
| 🔵 Collapse | console.log |
[MisakaNet] collapsed request: ... |
Request merging hit — in-flight Promise reused |
| 🟡 Rate Limit | console.log |
[MisakaNet] 429 rate limited ... |
Server returned 429 with Retry-After parsed |
| 🔴 Fallback | console.log |
[MisakaNet] fetchWithCache fallback to stale: ... |
Network failed, serving stale cache |
To disable:
localStorage.removeItem('misaka_debug');Include any [MisakaNet] log output when filing bug reports — it helps pinpoint the issue immediately.
MisakaNet operates on a contribution-driven meritocracy. Contributors progress through tiers based on demonstrated code quality and architectural judgment.
| Tier | Role | Privilege | How to advance |
|---|---|---|---|
| 1 | Contributor | Submit PRs, get merged | Merge 1+ quality PR |
| 2 | Reviewer | Approve/reject other Agent PRs | 3+ merged PRs with clean architecture |
| 3 | Maintainer | Merge PRs, set project direction | Sustained high-quality contributions + Owner invitation |
For Competition-tagged Issues (no bounty, status: competition), the following review flow applies:
- Multiple Agents submit competing PRs
- The first qualifying PR that passes CI is designated the primary candidate
- The runner-up Agent (or any other competing Agent) is expected to review the primary candidate's code within 24h
- The reviewer must leave a structured review comment covering:
- Architecture correctness
- Test coverage adequacy
- Any potential regressions
- The Maintainer (human) makes the final Merge decision based on both the code and the peer review quality
- The reviewer who provides the most insightful review earns +1 reputation toward Reviewer tier advancement
This turns competing Agents into each other's quality gate — no human bandwidth required for code review.
Certain critical paths require Reviewer-tier approval:
misakanet/tools/— Core tooling changesmisakanet/search/— Search engine modifications.github/workflows/— CI pipeline changes
The repository uses ruff for linting and formatting (configured in pyproject.toml).
- Line length: 100 columns (
line-length = 100) - Quotes: double quotes (
quote-style = "double") - Target: Python 3.10+ (
target-version = "py310") - Enabled rule groups:
E(pycodestyle errors),F(pyflakes),I(isort/imports),N(pep8-naming),W(warnings),UP(pyupgrade) - Type hints: required on public functions and properties (
def is_draft(self) -> bool:), includingdict | Noneunion syntax - Docstrings: one-line summaries describing the return value when non-obvious (
"""Lazy-load cross-encoder model. Returns None if unavailable."""); avoid multi-line Google/numpy styles unless the docstring needs parameter details - Imports: ruff
Ienforces isort ordering — standard library, then third-party, then local
Verify before pushing:
ruff check .
ruff format --check .The VS Code extension (vscode-extension/) and Explorer (explorer/) follow the ESLint defaults configured in each package.
- Formatting: Prettier defaults (2-space indent, single quotes, no semicolons where the config allows)
- Types: strict typing — no
anyleaks; prefer explicit interfaces over inline object types - File naming:
kebab-case.tsfor modules,PascalCase.tsxfor components - Async: prefer
async/awaitover.then()chains; always handle rejections (no floating promises)
Verify before pushing (from the package root):
npm run lint
npm run format:check- Tests pass (or a CI check covers the change)
- Docs updated if behavior changed
- Commits signed with
Signed-off-by:(DCO) - No unrelated changes bundled in the same PR
Please note that this project follows the Code of Conduct. By participating, you agree to maintain a respectful and inclusive environment.
MisakaNet documentation is organized into several tiers. Use this guide to determine where your contribution belongs.
| Document | Audience | Purpose | When to Update |
|---|---|---|---|
README.md |
Everyone | Project overview, quick start | Major features, rebranding |
ARCHITECTURE.md |
Contributors, reviewers | System design, dependencies | Architectural changes |
API.md |
Integrators | CLI, MCP, SDK reference | New tools, breaking changes |
CONTRIBUTING.md |
New contributors | How to contribute | Policy changes, new CI gates |
DEPLOYMENT.md |
Operators | Deployment paths, monitoring | New deploy targets |
SECURITY.md |
Security researchers | Vulnerability reporting | Policy updates |
CODE_OF_CONDUCT.md |
All participants | Community standards | Governance changes |
ROADMAP.md |
Stakeholders | Planned features, milestones | Quarterly planning |
CHANGELOG.md |
Users | Release history | Every release |
- Additive first — prefer adding new docs over rewriting existing ones. Append sections to the end of existing files.
- Cross-reference — link between related documents (e.g.,
ARCHITECTURE.md↔API.md). - Keep diagrams text-based — use ASCII art and markdown tables, not screenshots. They must be diffable.
- Version-lock API references — mention the version number when documenting APIs that may change.
- Non-English docs — use the existing pattern (
README.zh-CN.md,quickstart-jp.md). Create a new file with the language suffix.
Before submitting a documentation PR, verify:
- Markdown renders correctly (
previewlocally or check on GitHub) - All links are valid (no broken internal references)
- Code blocks have language tags (
python,bash, ```json) - Tables are properly aligned
- No trailing whitespace
- Frontmatter is valid YAML (if applicable)