[!IMPORTANT]
GitContribute never writes to GitHub. It syncs public or authenticated read-only
data, keeps research on your machine, and prepares local drafts for you to review.
Why GitContribute?
Finding an issue is easy. Finding a contribution that is useful, unclaimed,
appropriately scoped, and backed by evidence is the hard part.
GitContribute gives developers and coding agents a durable SQLite corpus of
repositories, issues, pull requests, reviews, code snapshots, and research
artifacts. Network access is always explicit. Once data is synced, search,
inspection, health analysis, dossiers, and investigations run entirely offline.
|
Capability |
What it gives you |
| π |
Typed offline search |
Search repositories, issues, PRs, threads, and indexed code with transparent ranking. |
| π‘ |
Contribution Radar |
Rank promising open issues with explicit positives, risks, blockers, confidence, and missing-evidence warnings. |
| ποΈ |
Durable research corpus |
Keep observations, coverage, investigations, evidence, and outcomes in local SQLite. |
| π§ |
Contribution workflow |
Move from hypothesis to opportunity, workspace, validation, readiness checks, and a prepared issue, PR, or review. |
| π€ |
Agent-ready MCP server |
Give Codex or Claude Code structured tools and resources with explicit capability boundaries. |
| π‘οΈ |
Safe by default |
Separate network reads, local writes, process execution, and GitHub mutation. |
β‘ Quick start
Run the guided setup with Node.js 18 or newer:
npx gitcontribute@latest setup
The interactive wizard offers two independent capabilities:
- a persistent
gitcontribute command for the CLI and TUI;
- MCP registration for Codex and/or Claude Code.
The wizard detects installed coding agents, preselects them in a visual
multi-select, and shows the exact configuration paths it will update. GitHub
authentication is chosen from described options rather than typed internal
identifiers. A dry-run plan still shows every effect before final confirmation.
Installing the terminal app is the recommended default. The wizard shows the
exact global npm command before applying changes. MCP-only setup remains
available and does not require a persistent CLI installation.
Then launch the TUI, sync a repository, rank contribution candidates, and
search the local corpus:
gitcontribute tui
gitcontribute sync owner/repo
gitcontribute radar owner/repo --limit 10
gitcontribute search threads "connection timeout" \
--repo owner/repo --json
gitcontribute dossier build owner/repo --json
setup initializes the corpus, helps select a GitHub authentication source,
can install the terminal app, and can register the MCP server with Codex and
Claude Code. Adding a repository during setup does not contact GitHub or
begin a sync.
For non-interactive setup, terminal installation must be explicitly selected:
# Recommended: terminal app plus Codex MCP
npx gitcontribute@latest setup --install-cli --codex --yes
# Terminal app only
npx gitcontribute@latest setup --install-cli --no-mcp --yes
# MCP only; leaves the terminal command uninstalled
npx gitcontribute@latest setup --codex --yes
When terminal installation is skipped, continue to prefix direct commands with
npx gitcontribute@latest.
Other installation options
Install a persistent command
npm install --global gitcontribute@latest
gitcontribute setup
To remove a global npm installation later, run
npm uninstall --global gitcontribute. The gitcontribute remove command only
removes MCP registrations; it does not uninstall the terminal app.
Pin a project version
npm install --save-dev gitcontribute
npx gitcontribute setup --codex --yes
The npm package has no install lifecycle and performs no download during
installation. Native binaries are included for macOS ARM64/x64, Linux ARM64/x64,
and Windows x64.
Build from source
Developers with Go 1.26 or newer can install or build the CLI directly:
go install github.com/morluto/gitcontribute/cmd/gitcontribute@latest
go build -o gitcontribute ./cmd/gitcontribute
You need git. The gh CLI is optional and can provide authentication through
gh auth token; ripgrep is optional and adds an extra doctor check.
π§© How it works
GitHub read APIs Local checkout
β β
β explicit sync / hydrate β explicit index / acquire
βΌ βΌ
ββββββββββββββββββββββββββββββββββββββββββββ
β Local SQLite corpus β
β observations Β· projections Β· coverage β
β investigations Β· evidence Β· outcomes β
ββββββββββββββββββββββ¬ββββββββββββββββββββββ
β offline reads
ββββββββββββββΌββββββββββββββ
βΌ βΌ βΌ
CLI / TUI MCP agents JSON exports
1. Discover
Track explicit repositories, GitHub search results, or GH Archive event streams.
Sync only when you ask, then search locally as often as you like.
gitcontribute source add repos --name my-go "golang/go" "cli/cli" --json
gitcontribute crawl my-go --since 720h --budget 500 --json
gitcontribute radar golang/go --limit 20 --json
gitcontribute search issues "data race" --repo golang/go --state open --json
gitcontribute search code "context.WithTimeout" --repo golang/go
2. Investigate
Build a dossier or per-thread research brief, inspect repository health, record
a hypothesis, and check for duplicate or competing work before committing time.
gitcontribute dossier build owner/repo
gitcontribute research brief issue:owner/repo#42 --format markdown
gitcontribute health owner/repo --json
gitcontribute investigation start-thread issue:owner/repo#42 --json
gitcontribute duplicates check <hypothesis-id>
gitcontribute collisions check <hypothesis-id>
3. Validate
Promote promising research into an opportunity, create an isolated worktree,
record evidence, and compare a baseline with your candidate change.
gitcontribute opportunity promote --problem="Retry can hang indefinitely" \
--scope=small --impact="reduces flakes" --effort=1h \
--confidence=0.8 <hypothesis-id>
gitcontribute workspace create <investigation-id>
gitcontribute validation define --kind=test --command="go test ./..." \
--working-dir=/path/to/workspace <investigation-id>
gitcontribute validation run <validation-id> --kind=base --execute
gitcontribute validation run <validation-id> --kind=candidate --execute
gitcontribute validation compare <base-run-id> <candidate-run-id>
4. Check readiness
Run a deterministic local readiness gate before turning a candidate into a
public submission:
gitcontribute readiness opportunity <opportunity-id>
gitcontribute readiness explain <opportunity-id>:evidence_freshness
Readiness reports pass, warn, block, or unknown per rule. It only reads
stored corpus and draft metadata: it does not fetch GitHub, execute validation,
write state, or infer that missing coverage is a hard blocker.
5. Prepare
Create a local contribution draft from the evidence and workspace diff:
gitcontribute prepare issue <opportunity-id>
gitcontribute prepare pr --approach="Bound retries with context" \
--workspace <workspace-id> <opportunity-id>
gitcontribute prepare review <opportunity-id>
Nothing is posted. You decide what leaves your machine.
π€ Use with AI agents
The MCP server gives agents structured access to the same corpus and workflow:
gitcontribute setup --codex --yes
gitcontribute setup --all-clients --yes
Or start the stdio server directly:
gitcontribute mcp serve --transport=stdio
MCP capabilities are deliberately separate:
Tool names use the gitcontribute.<capability>.<action> namespace. The server
advertises one canonical name per operation; unnamespaced compatibility aliases
are not registered.
| Capability |
Examples |
| Offline reads |
Search, inspect repositories and threads, build research briefs, read dossiers, explain matches, inspect evidence, opportunities, readiness checks, and workflow resources. |
| Network reads |
Sync repositories, hydrate threads, start crawls, and acquire workspaces. |
| Local writes |
Start investigations, record hypotheses, promote opportunities, define validations, and prepare drafts. |
| Execution |
Run a validation only when the request includes execute: true. |
Contribution workflow resources and prompts are available for agents:
gitcontribute.corpus.get_readiness and gitcontribute://readiness/<opportunity-id> expose the
same offline readiness report as the CLI.
gitcontribute://workflow/contribution/<opportunity-id> links the local
opportunity, evidence, readiness report, and safe workflow prompts.
- Prompts such as
review_contribution_readiness instruct agents to treat
repository and GitHub-sourced text as untrusted data and to ask before any
network read, local write, or validation execution.
Resources are published under gitcontribute:// and github-index:// URI
schemes. See the architecture guide for the complete
application and adapter boundaries.
π‘οΈ Safety by design
| Operation |
Network |
Local write |
Runs a process |
GitHub write |
| Search, health, dossier, research-brief, and readiness inspection |
β |
β |
β |
β |
| Investigations, evidence, lenses |
β |
β |
β |
β |
| Sync, crawl, hydrate |
β |
β |
β |
β |
| Acquire or create a workspace |
remote-dependent |
β |
git only |
β |
| Validation with explicit execution |
β by default |
β |
β |
β |
- No GitHub writes. GitContribute does not open issues, create pull requests,
push commits, or mutate GitHub.
- No hidden network access. Corpus reads never fetch data.
- No hosted service or telemetry. Your corpus and research remain local.
- No automatic repository execution. Crawling and indexing never execute
repository-controlled code.
- No implied sandbox. Explicit validation commands run on your host with the
permissions of your user and only the environment variables you allowlist.
- No opaque semantic ranking. Search uses SQLite FTS5 and reports signals
such as text matches, freshness, and coverage.
π CLI guide
The sections below are a task-oriented reference. Run gitcontribute --help or
gitcontribute <command> --help for every flag.
Setup, configuration, and authentication
gitcontribute setup # interactive
gitcontribute setup --codex --yes # configure Codex
gitcontribute setup --all-clients --yes # configure supported clients
gitcontribute setup --codex --mcp-version latest --yes
gitcontribute setup --token-source env \
--token-source-key GITHUB_TOKEN --yes
gitcontribute setup --codex --dry-run --json # inspect without writing
gitcontribute remove --all-clients --yes # remove MCP registrations only
gitcontribute upgrade --check
gitcontribute upgrade --yes
gitcontribute init creates the default database and directories without
contacting GitHub. gitcontribute configure updates config.toml atomically:
gitcontribute configure --database /path/to/corpus.db
gitcontribute configure --token-source env --token-source-key GITHUB_TOKEN
gitcontribute configure --token-source gh-cli
gitcontribute configure --output-format json
Authentication sources are none, env, gh-cli, and keyring. Tokens are
resolved at runtime and are never stored in the corpus or logs. Use
gitcontribute status, metadata, and doctor to inspect the local setup.
See the onboarding design for the full contract and
environment-variable reference.
Sources, crawling, sync, and hydration
Add a source:
gitcontribute source add repos --name my-go "golang/go" "cli/cli" --json
gitcontribute source add search --name go-network \
--query "language:go stars:>100" --json
gitcontribute source add gharchive --name golang-events \
--events "IssuesEvent,PullRequestEvent" --json
Run it once or continuously:
gitcontribute crawl golang-events --since 720h --budget 500 --json
gitcontribute tail golang-events --since 2h --budget 500 --interval 1h
Sync and selectively hydrate repository archives:
gitcontribute sync owner/repo
gitcontribute archive sync owner/repo --since 168h --state open
gitcontribute archive sync owner/repo --numbers 42,108
gitcontribute archive refresh owner/repo
gitcontribute archive hydrate owner/repo#42 --with issue_comments
gitcontribute archive hydrate owner/repo#108 \
--with pr_reviews,pr_review_comments
gitcontribute archive coverage owner/repo
Hydration supports issue_comments, pr_details, pr_reviews, and
pr_review_comments. Fetches are paginated and cancellation-aware; a complete
facet replaces its previous snapshot atomically.
Code indexing and acquisition
Index a clean local checkout at its current commit:
gitcontribute index owner/repo /path/to/checkout --json
Or acquire a repository into a managed mirror, index a clean temporary
worktree, and remove that worktree afterward:
gitcontribute acquire owner/repo \
--remote https://github.com/owner/repo.git --json
The indexer reads blobs directly from Git, skips binaries and non-UTF-8 content,
enforces size limits, and rejects dirty worktrees.
Radar, research briefs, search, dossiers, health, seeds, and lenses
gitcontribute radar owner/repo --limit 20
gitcontribute search repos "cli" --limit 20 --json
gitcontribute search issues "data race" --repo owner/repo --state open --json
gitcontribute search prs "flaky" --repo owner/repo --label bug --json
gitcontribute search threads "memory leak" --repo owner/repo
gitcontribute search code "context.WithTimeout" --repo owner/repo
gitcontribute search all "retry" --repo owner/repo
gitcontribute dossier build owner/repo
gitcontribute dossier export owner/repo --format markdown \
--output owner-repo-dossier.md
gitcontribute research brief owner/repo#42
gitcontribute research brief pr:owner/repo#108 --json
gitcontribute health owner/repo --stale-after 336h --json
gitcontribute seeds owner/repo --json
radar is a strict offline corpus read. It ranks a bounded population of the
newest stored open issues and separates objective eligibility from the numeric
score. Every candidate reports positive signals, risks, blockers, confidence,
linked open PRs, evidence coverage, and unknowns. Missing comment coverage is
reported as unknown and does not lower the score; hydrate comments explicitly
when you want that evidence:
gitcontribute archive hydrate owner/repo#42 --with issue_comments
gitcontribute radar owner/repo --json
Radar scores carry a version (radar.v1) so saved JSON remains auditable when
ranking semantics evolve. It never syncs, hydrates, executes repository code,
or writes to GitHub.
research brief is also a strict offline read. Its versioned
research-brief.v1 output has fixed sections for state, stored problem fields,
acceptance hints, participants, timeline, explicit/duplicate references,
linked PRs, indexed code, contribution guidance, health, coverage gaps, and
next commands. Every section carries source references or an explicit unknown
reason. Checkboxes and maintainer phrases remain source extractsβnot fabricated
or complete acceptance criteria. Markdown output redacts common credentials and
quotes untrusted source text; JSON ordering is deterministic.
Missing child facets and code stay visible instead of triggering hidden work:
gitcontribute archive hydrate owner/repo#42 --with issue_comments
gitcontribute index owner/repo /path/to/clean-checkout
gitcontribute research brief issue:owner/repo#42 --json
Use a lens to apply saved filters and weighted ranking to a bounded population:
gitcontribute lens add my-lens --file lens.json
gitcontribute search issues "retry" --lens my-lens
gitcontribute lens explain my-lens issue:owner/repo#42 --query "retry"
Search results explain their scores. Most typed searches support opaque cursor
pagination; search all and lens-ranked searches do not.
Investigations, evidence, tracking, and collections
Start an investigation and its initial hypothesis from one exact stored thread
revision:
gitcontribute investigation start-thread issue:owner/repo#42 --json
This is a local-write operation with no network or process execution. The
investigation saves the immutable observation ID, source update time, sequence,
and source reference used for its title and bounded description. Repeating the
command returns the existing open pair with "created": false; it never updates
that baseline silently. The seed category is other because the command does
not infer scope or defect type from untrusted issue text.
The manual two-command path remains available when no stored thread should be
the baseline:
gitcontribute investigation start owner/repo --json
gitcontribute hypothesis add --title="Fix retry timeout" \
--description="Reproduce and isolate the timeout." \
--category=bug <investigation-id>
Record supporting or contradicting evidence:
gitcontribute evidence add --type=manual_observation --relation=supporting \
--description="Reproduced on the current default branch." \
--opportunity <opportunity-id>
Evidence shown through the CLI, exports, and MCP includes a derived freshness
status. github_source evidence recorded from a started thread carries the
exact corpus source revision it used, so later issue, pull request, facet, or
guidance updates can mark it stale. Missing source projections are
unknown; local/manual evidence is not_applicable. Freshness is evaluated
from local corpus reads only, and stale evidence is a prompt to re-check, not a
claim that the evidence is invalid.
Group typed references and record local decisions:
gitcontribute collection create interesting
gitcontribute collection add interesting \
repo:owner/repo issue:owner/repo#42
gitcontribute triage record issue:owner/repo#42 viewed --reason "..."
gitcontribute contribution record <opportunity-id> issue \
"Draft title" --body "..."
gitcontribute contribution outcome <contribution-id> submitted
Export and restore tracking metadata:
gitcontribute tracking export --output tracking.json
gitcontribute tracking import --file tracking.json
Tracking exports use schema version 2 for portable evidence provenance and
remain backward compatible with older unversioned tracking bundles. Exports are
redacted: they exclude credentials, tokens, and absolute local paths.
Workspaces, jobs, runs, and TUI
gitcontribute workspace show <workspace-id>
gitcontribute diff <workspace-id>
gitcontribute runs --limit 20
gitcontribute jobs
gitcontribute job show <id>
gitcontribute job cancel <id>
gitcontribute tui owner/repo
diff returns the patch, changed files, and suggested review order. The TUI is
local-only; add --json to emit a non-interactive snapshot.
JSON and output behavior
Most non-interactive commands accept --json. Machine-readable output goes to
stdout; progress and status messages go to stderr. List commands accept
--limit, and paginated searches return an opaque next_cursor where supported.
dossier export, export dossier, export evidence, and tracking export
accept --output <file>.
πΎ Storage locations
| Platform |
Config |
Data |
Cache |
Logs |
| Linux / Unix |
$XDG_CONFIG_HOME/gitcontribute or ~/.config/gitcontribute |
$XDG_DATA_HOME/gitcontribute or ~/.local/share/gitcontribute |
$XDG_CACHE_HOME/gitcontribute or ~/.cache/gitcontribute |
$XDG_STATE_HOME/gitcontribute or ~/.local/state/gitcontribute/logs |
| macOS |
~/Library/Application Support/gitcontribute |
~/Library/Application Support/gitcontribute/Data |
~/Library/Caches/gitcontribute |
~/Library/Logs/gitcontribute |
| Windows |
%APPDATA%\gitcontribute |
%LOCALAPPDATA%\gitcontribute\Data |
%LOCALAPPDATA%\gitcontribute\Cache |
%LOCALAPPDATA%\gitcontribute\Logs |
The default database is gitcontribute.db in the data directory. The
configuration file is config.toml in the config directory.
Linux and macOS are the primary development and test targets. Windows builds
are expected to work with Git for Windows and standard %APPDATA% and
%LOCALAPPDATA% paths. For platform-specific problems, open a bug report with
the output of gitcontribute doctor --json.
π οΈ Development
go test ./...
go build -o gitcontribute ./cmd/gitcontribute
./gitcontribute --help
Before changing package boundaries or side effects, read
docs/architecture.md. See CONTRIBUTING.md
for the complete development and testing workflow.