adnanmokhtar/refract
Write your AI coding setup once, run it in every tool. One source of truth — 184 commands, 99 agents, 234 skills, 24 role-based packs — compiled into the native format of Claude Code, Cursor, Gemini CLI, OpenCode, Aider, Cline, Codex, Conti
field: unknown workflow: Testing & TDDData / ETLSecurity reviewCI/CD & deployPlanning & specs runs claude in CI
ARCHETYPE
Fort Knox
Deny lists, pre-tool hooks, sandboxing. Nothing touches prod without a signature.
GUARDRAILS
4/5
Blocks destructive commands · Pre-tool screening hook · No YOLO mode · Sandbox or ask-first rules · details
Quality 74/100 Guardrails v2
Copy this rig
# review before running: this installs third-party code
$ npx degit adnanmokhtar/refract/templates/repo-baseline/.claude ./rig-refract # inspect, then merge into .claude/ MCP servers are added to Claude Code at local scope; env vars are shown as YOUR_… placeholders — we never store values. Files are fetched with degit into a separate folder so you can review before merging.
$ npx degit adnanmokhtar/refract/templates/packs/ai-engineering/skills/eval-run .claude/skills/eval-run $ npx degit adnanmokhtar/refract/templates/packs/ai-engineering/skills/llm-gateway-audit .claude/skills/llm-gateway-audit $ npx degit adnanmokhtar/refract/templates/packs/ai-engineering/skills/prompt-audit .claude/skills/prompt-audit $ npx degit adnanmokhtar/refract/templates/packs/ai-engineering/skills/retrieval-eval .claude/skills/retrieval-eval $ npx degit adnanmokhtar/refract/templates/packs/ai-engineering/skills/vector-index-audit .claude/skills/vector-index-audit $ npx degit <redacted>-contract-scan .claude/skills/algorithm-contract-scan $ npx degit <redacted>-derivation .claude/skills/complexity-derivation $ npx degit <redacted>-scan .claude/skills/complexity-scan $ npx degit <redacted>-stability-scan .claude/skills/numeric-stability-scan $ npx degit <redacted>-detect .claude/skills/align-detect $ npx degit <redacted>-oracle .claude/skills/convention-oracle $ npx degit <redacted>-ratchet-scan .claude/skills/convention-ratchet-scan $ npx degit <redacted>-and-align .claude/skills/find-and-align $ npx degit <redacted>-layout-scan .claude/skills/naming-layout-scan $ npx degit <redacted>-adoption-scan .claude/skills/wrapper-adoption-scan $ npx degit <redacted>-consistency-audit .claude/skills/api-consistency-audit $ npx degit <redacted>-snapshot .claude/skills/api-snapshot $ npx degit <redacted>-tenant .claude/skills/debug-tenant $ npx degit <redacted>-error-scan .claude/skills/edge-error-scan $ npx degit <redacted>-test-method .claude/skills/endpoint-test-method $ npx degit <redacted>-diff .claude/skills/env-diff $ npx degit <redacted>-contract-scan .claude/skills/http-contract-scan $ npx degit <redacted>-lifecycle-scan .claude/skills/job-lifecycle-scan $ npx degit <redacted>-tail-method .claude/skills/log-tail-method $ npx degit <redacted>-scaffold .claude/skills/module-scaffold $ npx degit <redacted>-independent-ops .claude/skills/parallelize-independent-ops $ npx degit <redacted>-transition-scan .claude/skills/state-transition-scan $ npx degit <redacted>-boundary-scan .claude/skills/validation-boundary-scan $ npx degit <redacted>-format-scan .claude/skills/wire-format-scan $ npx degit <redacted>-business-coverage .claude/skills/check-business-coverage $ npx degit <redacted>-scan .claude/skills/ledger-scan $ npx degit <redacted>-scan .claude/skills/lifecycle-scan $ npx degit <redacted>-rule-scan .claude/skills/pricing-rule-scan $ npx degit <redacted>-scan .claude/skills/report-scan $ npx degit <redacted>-invoice-scan .claude/skills/tax-invoice-scan $ npx degit adnanmokhtar/refract/templates/packs/code-quality/skills/architectural-diagnosis .claude/skills/architectural-diagnosis $ npx degit adnanmokhtar/refract/templates/packs/code-quality/skills/change-brief .claude/skills/change-brief $ npx degit adnanmokhtar/refract/templates/packs/code-quality/skills/correctness-smell-scan .claude/skills/correctness-smell-scan $ npx degit adnanmokhtar/refract/templates/packs/code-quality/skills/dead-code-scan .claude/skills/dead-code-scan $ npx degit adnanmokhtar/refract/templates/packs/code-quality/skills/debt-ledger .claude/skills/debt-ledger
$ curl -fsSL --create-dirs -o .claude/agents/test-engineer.md https://raw.githubusercontent.<redacted>-stale-knowledge-slice/resources/.claude/agents/test-engineer.md $ curl -fsSL --create-dirs -o .claude/agents/test-engineer.md https://raw.githubusercontent.<redacted>-stale-knowledge-slice/resources/.claude/agents/test-engineer.md $ curl -fsSL --create-dirs -o .claude/agents/verifier.md https://raw.githubusercontent.com/adnanmokhtar/refract/main/templates/repo-baseline/.claude/agents/verifier.md $ curl -fsSL --create-dirs -o .claude/agents/project-dispatcher.md https://raw.githubusercontent.<redacted>-baseline/.claude/agents/project-dispatcher.md
Merge into .claude/settings.json (project) or ~/.claude/settings.json (user). Hook commands reference scripts in the source repo — copy those too.
{
"permissions": {
"deny": [
"Bash(rm -rf:*)",
"Bash(rm -fr:*)",
"Bash(git push --force:*)",
"Bash(git push -f:*)",
"Bash(git reset --hard:*)",
"Bash(git clean -fd:*)",
"Bash(git branch -D:*)",
"Bash(git commit --amend:*)",
"Bash(*uninstall*)",
"Bash(*:--no-verify:*)",
"Bash(rm -rf:*)",
"Bash(rm -fr:*)",
"Bash(git push --force:*)",
"Bash(git push -f:*)",
"Bash(git reset --hard:*)",
"Bash(git clean -fd:*)",
"Bash(git branch -D:*)"
]
},
"hooks": {
"PreToolUse": [
{
"matcher": "Edit|Write|MultiEdit",
"hooks": [
{
"type": "command",
"command": "cd \"${CLAUDE_PROJECT_DIR:-.}\" && .claude/hooks/test-lock.sh"
},
{
"type": "command",
"command": "cd \"${CLAUDE_PROJECT_DIR:-.}\" && .claude/hooks/artifact-ownership.sh"
},
{
"type": "command",
"command": "cd \"${CLAUDE_PROJECT_DIR:-.}\" && .claude/hooks/pre-edit-guard.sh"
},
{
"type": "command",
"command": "cd \"${CLAUDE_PROJECT_DIR:-.}\" && .claude/hooks/secret-scan.sh"
},
{
"type": "command",
"command": "cd \"${CLAUDE_PROJECT_DIR:-.}\" && .claude/hooks/module-boundaries.sh"
},
{
"type": "command",
"command": "cd \"${CLAUDE_PROJECT_DIR:-.}\" && .claude/hooks/inject-path-rules.sh"
},
{
"type": "command",
"command": "cd \"${CLAUDE_PROJECT_DIR:-.}\" && .claude/hooks/inject-blast-radius.sh"
}
]
},
{
"matcher": "Bash",
"hooks": [
{
"type": "command",
"command": "cd \"${CLAUDE_PROJECT_DIR:-.}\" && .claude/hooks/guard-destructive.sh"
},
{
"type": "command",
"command": "cd \"${CLAUDE_PROJECT_DIR:-.}\" && .claude/hooks/guard-irreversible.sh"
},
{
"type": "command",
"command": "cd \"${CLAUDE_PROJECT_DIR:-.}\" && .claude/hooks/test-lane.sh"
},
{
"type": "command",
"command": "cd \"${CLAUDE_PROJECT_DIR:-.}\" && .claude/hooks/test-lock.sh"
},
{
"type": "command",
"command": "cd \"${CLAUDE_PROJECT_DIR:-.}\" && .claude/hooks/judge-bash-guard.sh"
}
]
},
{
"matcher": "Read|Grep|Glob|Agent|Task",
"hooks": [
{
"type": "command",
"command": "cd \"${CLAUDE_PROJECT_DIR:-.}\" && .claude/hooks/judge-bash-guard.sh"
},
{
"type": "command",
"command": "cd \"${CLAUDE_PROJECT_DIR:-.}\" && .claude/hooks/artifact-ownership.sh"
}
]
}
]
},
"sandbox": {
"enabled": true
}
} Skills (148)
eval-runllm-gateway-auditprompt-auditretrieval-evalvector-index-auditalgorithm-contract-scancomplexity-derivationcomplexity-scannumeric-stability-scanalign-detectconvention-oracleconvention-ratchet-scanfind-and-alignnaming-layout-scanwrapper-adoption-scanapi-consistency-auditapi-snapshotdebug-tenantedge-error-scanendpoint-test-methodenv-diffhttp-contract-scanjob-lifecycle-scanlog-tail-methodmodule-scaffoldparallelize-independent-opsstate-transition-scanvalidation-boundary-scanwire-format-scancheck-business-coverageledger-scanlifecycle-scanpricing-rule-scanreport-scantax-invoice-scanarchitectural-diagnosischange-briefcorrectness-smell-scandead-code-scandebt-ledgerduplication-scanerror-handling-scanrefactoring-sweepsmoke-verifysuppression-scantest-shieldtype-safety-scancontract-diffdata-test-scanderived-data-scanextract-scangrain-probeincremental-load-scanlineage-traceorchestration-scanpii-copy-scanwarehouse-scan-auditconnection-config-scanindex-coverage-scanmigration-rehearsalmigration-safetyn-plus-one-scanquery-shape-scanschema-consistency-auditschema-diffschema-integrity-scantransaction-scanlane-backendlane-frontendlane-mobilecompose-scancontainer-build-scandeploy-safety-scangitops-auditmonitor-deploypipeline-cache-scanpipeline-integrity-scanprogressive-deliveryrelease-integrity-scanrelease-securitychaos-testcoordination-scandlq-replaydual-write-scanevent-contract-scanmessaging-scanretry-deadline-scanworkflow-scanchangelog-generatediagram-syncdoc-drift-scandoc-sample-scandoc-structure-scandocstring-coveragequickstart-verifyallocation-scancandidate-pricingcommitment-coverageegress-traceguardrail-scanspend-anomaly-triagespend-path-scanstorage-growth-scanunit-cost-probea11y-code-scana11y-scanbundle-analyzeclient-data-scancomponent-playgroundcritical-path-scandev-server-startfont-optimizationi18n-scanimage-optimizationlcp-auditlighthouse-cinavigation-speedrender-correctness-scanrouting-scanseo-auditssr-auditstreaming-ssrverify-with-playwrightvisual-checkadmission-policydns-tls-lifecycle-scandr-auditedge-reliability-scaniac-cost-scaniac-lifecycle-scaniac-state-scank8s-api-version-scank8s-auditk8s-availability-scank8s-render-scank8s-wiring-scannetwork-exposure-auditserverless-reliability-scantf-plan-reviewapply-pack-adaptationcompute-anchor-densitycontext-load-scandecision-record-scanextract-architecture-deeplyextract-base-class-idiomextract-business-contextextract-codebase-overviewextract-conventions-emerging
Subagents (4)
| test-engineer | Writes the failing test first and proves the fix with it. |
| test-engineer | Writes the failing test first and proves the fix with it. |
| verifier model: opus | Independent fresh-context judge, dispatched by pipeline commands with a brief: grades a run's locked acceptance contract with tests it runs at HEAD (mode acceptance), or refutes candidate review findi |
| project-dispatcher | Reads PROJECTS.md and determines which sibling repos a task affects. Produces impact matrices for cross-repo work. |
Hooks (27)
| event | matcher | runs |
|---|---|---|
| PostToolUse | Edit|Write|MultiEdit | cd "${CLAUDE_PROJECT_DIR:-.}" && .claude/hooks/post-edit-check.sh |
| PostToolUse | Edit|Write|MultiEdit | cd "${CLAUDE_PROJECT_DIR:-.}" && .claude/hooks/format-on-save.sh |
| PostToolUse | Edit|Write|MultiEdit | cd "${CLAUDE_PROJECT_DIR:-.}" && .claude/hooks/auto-test.sh |
| PreToolUse | Edit|Write|MultiEdit | cd "${CLAUDE_PROJECT_DIR:-.}" && .claude/hooks/test-lock.sh |
| PreToolUse | Edit|Write|MultiEdit | cd "${CLAUDE_PROJECT_DIR:-.}" && .claude/hooks/artifact-ownership.sh |
| PreToolUse | Edit|Write|MultiEdit | cd "${CLAUDE_PROJECT_DIR:-.}" && .claude/hooks/pre-edit-guard.sh |
| PreToolUse | Edit|Write|MultiEdit | cd "${CLAUDE_PROJECT_DIR:-.}" && .claude/hooks/secret-scan.sh |
| PreToolUse | Edit|Write|MultiEdit | cd "${CLAUDE_PROJECT_DIR:-.}" && .claude/hooks/module-boundaries.sh |
| PreToolUse | Edit|Write|MultiEdit | cd "${CLAUDE_PROJECT_DIR:-.}" && .claude/hooks/inject-path-rules.sh |
| PreToolUse | Edit|Write|MultiEdit | cd "${CLAUDE_PROJECT_DIR:-.}" && .claude/hooks/inject-blast-radius.sh |
| PreToolUse | Bash | cd "${CLAUDE_PROJECT_DIR:-.}" && .claude/hooks/guard-destructive.sh |
| PreToolUse | Bash | cd "${CLAUDE_PROJECT_DIR:-.}" && .claude/hooks/guard-irreversible.sh |
| PreToolUse | Bash | cd "${CLAUDE_PROJECT_DIR:-.}" && .claude/hooks/test-lane.sh |
| PreToolUse | Bash | cd "${CLAUDE_PROJECT_DIR:-.}" && .claude/hooks/test-lock.sh |
| PreToolUse | Bash | cd "${CLAUDE_PROJECT_DIR:-.}" && .claude/hooks/judge-bash-guard.sh |
| PreToolUse | Read|Grep|Glob|Agent|Task | cd "${CLAUDE_PROJECT_DIR:-.}" && .claude/hooks/judge-bash-guard.sh |
| PreToolUse | Read|Grep|Glob|Agent|Task | cd "${CLAUDE_PROJECT_DIR:-.}" && .claude/hooks/artifact-ownership.sh |
| UserPromptSubmit | * | cd "${CLAUDE_PROJECT_DIR:-.}" && .claude/hooks/recall-inject.sh |
| SessionStart | * | cd "${CLAUDE_PROJECT_DIR:-.}" && .claude/hooks/session-start.sh |
| PreCompact | * | cd "${CLAUDE_PROJECT_DIR:-.}" && .claude/hooks/precompact-handoff.sh |
| Stop | * | cd "${CLAUDE_PROJECT_DIR:-.}" && .claude/hooks/update-session-log.sh |
| Stop | * | cd "${CLAUDE_PROJECT_DIR:-.}" && .claude/hooks/verify-gate.sh |
| SubagentStop | ^(.+:)?verifier$ | cd "${CLAUDE_PROJECT_DIR:-.}" && .claude/hooks/persist-verdict.sh |
| SubagentStop | * | cd "${CLAUDE_PROJECT_DIR:-.}" && .claude/hooks/persist-review.sh |
| Notification | * | cd "${CLAUDE_PROJECT_DIR:-.}" && .claude/hooks/notify.sh |
| PostToolUse | Edit|Write|MultiEdit | .claude/hooks/dispatch-hook.sh |
| SessionStart | * | .claude/hooks/session-start.sh |
Slash commands (10)
/catchup/execute-plan/fix-bug/ship/verify-plan/cross-repo-task/migration-doctor/migration-workspace-status/project-map/sync-contract
Permissions
deny (17)
Bash(rm -rf:*)
Bash(rm -fr:*)
Bash(git push --force:*)
Bash(git push -f:*)
Bash(git reset --hard:*)
Bash(git clean -fd:*)
Bash(git branch -D:*)
Bash(git commit --amend:*)
Bash(*uninstall*)
Bash(*:--no-verify:*)
Bash(rm -rf:*)
Bash(rm -fr:*)
Bash(git push --force:*)
Bash(git push -f:*)
Bash(git reset --hard:*)
Bash(git clean -fd:*)
Bash(git branch -D:*)
ask (0)
—
allow (14)
Bash(git status:*)
Bash(git diff:*)
Bash(git log:*)
Bash(git show:*)
Bash(git branch:*)
Bash(git rev-parse:*)
Bash(ls:*)
Bash(find:*)
Bash(grep:*)
Bash(rg:*)
Bash(cat:*)
Bash(head:*)
Bash(tail:*)
Bash(wc:*)
Similar rigs
kapadias/nonna
Your AI agent says done. Nonna makes it prove it: she runs your test suite before a coding agent can stop, and blocks pushes to main and secrets in files. Claude Code plugin, plus git hooks for Codex, Cursor, Copilot, Gemini and more
Fort Knox 3.9k tok · · Q78
Yeachan-Heo/oh-my-claudecode
Teams-first Multi-agent orchestration for Claude Code
Orchestrator 14.5k tok · · Q3
rajib2k5/claude-staff-kit
Staff-level engineering standards for Claude Code: skills, model-routed agents, commands, rules, contexts, and a destructive-ops guard. Installs as a plugin or as ~/.claude dotfiles.
Orchestrator 6.1k tok · · Q0
Fibteq-Inc/Myrel
Open-source, self-hostable AI workspace: a coordinator agent turns requests into threads, agents do the work in a cloud sandbox or with your own Claude Code, Codex, Antigravity CLI or OpenCode. Runs on Cloudflare.
Minimalist 260 tok · · Q0