Persistent memory and receipt-verified workflows for Claude Code. Your agent stops starting from zero, and its word stops being the evidence. 28 auto-triggering skills, zero commands: say what you need and Claude picks the right skill.
Version: 16.4.0 | License: Apache 2.0
Fix Receipts (
/fix) — new in 11.2.0. Sayfix this and prove it. Attune previews the contract first — done conditions, scope, probes, nothing executes — then--runreturns a receipt: the diff attributed against a pre-run snapshot, your probes re-run independently of the agent, and an exit code that means the probes passed — not that the agent felt good about it. The agent doesn't grade its own homework.
Spec Ladders (
/spec) — goal-driven development you approve rung by rung. Saybuild X. Attune interviews you until the goal becomes requirements, design, and a gated task ladder — then climbs it only as fast as you approve: every task behind your explicit go, every quality gate scored, every ruling recorded in a decision file that outlives the session. Between your approvals, the agent does the climbing — dispatching attune workflows and its own native tools. The spec isn't a plan you wrote once; it's a contract that drives the work.
claude plugin marketplace add Smart-AI-Memory/attune-ai
claude plugin install attune-ai@attune-aiUpdate an existing install:
claude plugin update attune-ai@attune-aiJust describe what you need in Claude Code:
- "what can attune do?" — triggers
attune-hub - "review my code" — triggers
code-quality - "scan for security issues" — triggers
security-audit - "generate tests for src/" — triggers
smart-test - "fix failing tests" — triggers
fix-test - "predict bugs" — triggers
bug-predict - "generate docs" — triggers
doc-gen - "plan this feature" — triggers
planning - "refactor this module" — triggers
refactor-plan - "prepare a release" — triggers
release-prep - "build from a spec" — triggers
spec
| Skill | Triggers On |
|---|---|
attune-hub |
what can attune do, help, capabilities |
spec |
spec ladders, build from scratch, brainstorm and execute |
security-audit |
security, vulnerability, scan |
code-quality |
review, quality, bugs, code smell |
bug-predict |
predict bugs, risky code, what might break |
doc-gen |
generate docs, documentation, README |
smart-test |
test gaps, generate tests, coverage |
fix-test |
fix test, broken test, debug test |
workflow-orchestration |
workflow, analyze, run |
planning |
plan, feature, architecture, TDD |
refactor-plan |
refactor, tech debt, simplify |
release-prep |
release, publish, deploy |
memory-and-context |
memory, store, retrieve |
Autonomous sub-agents (auto-discovered from plugin/agents/):
| Agent | Use it when |
|---|---|
setup-guide |
checking prerequisites; installing/configuring attune-ai, Redis, MCP |
spec-author |
spec a new feature — runs the SDD requirements interview and writes requirements.md (Phase 1 only; design/tasks stay gated) |
help-content-explainer |
explain an attune-help template for your repo — grounds the template's guidance in your actual code (read-only) |
security-reviewer |
read-only security pass — scans for eval/exec, path traversal, injection, secrets; reports findings by severity |
release-prep-auditor |
pre-release pre-flight — version/tree/CI/changelog/security/deps → ready or not-ready verdict (reports only) |
refactor-planner |
analyze a target for smells/duplication/complexity → prioritized refactoring roadmap (plans only) |
The plugin ships two security hooks:
- PreToolUse — blocks
eval(),exec(), path traversal, andrm -rf /in Bash; validates file paths in Edit/Write - PostToolUse — auto-formats Python with
blackandruff --fixafter Write/Edit
The plugin works standalone. Add the Python package for CLI automation, 66 MCP tools, multi-agent workflows, and cost tracking:
pip install 'attune-ai[developer]'| Capability | Plugin only | + pip |
|---|---|---|
| 28 auto-triggering skills | Yes | Yes |
| Prompt-based analysis | Yes | Yes |
| 66 MCP tools | -- | Yes |
attune CLI |
-- | Yes |
| Multi-agent workflows | -- | Yes |
| Cost tracking | -- | Yes |