claude-sentient
CLAUDE.md2 starsChanged 7 months agoMITView on GitHub
Project: Claude Sentient Version: 1.7.1 Type: Autonomous Development Orchestration Layer Claude Sentient is an orchestration layer that leverages Claude Code's native capabilities to work cohesively. It's not a replacement for Claude Code — it's a thin coordination layer that makes Claude Code's built-in features work together autonomously. When you invoke /cs-loop, Claude Sentient orchestrates: Detailed com
In this repository6 other files it gives its agents.
- .claude/commands/CLAUDE.mdCLAUDE.md
- .claude/hooks/CLAUDE.mdCLAUDE.md
- .cursor/rules/claude-sentient.mdcCursor rule
- profile-detectionSkill
- quality-gatesSkill
- team-orchestrationSkill
The file
We split this CLAUDE.md into the rules it gives an agent
2 rules. About 0 are common, 1 is distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 51
When you invoke /cs-loop, Claude Sentient orchestrates:
- Distinctive: line 113
Before committing, these must pass:
≈ “Common” is an estimate: at least 1 in 5 of the 28,029 CLAUDE.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
CLAUDE.md — Claude Sentient
Project: Claude Sentient Version: 1.7.1 Type: Autonomous Development Orchestration Layer
What Is This?
Claude Sentient is an orchestration layer that leverages Claude Code's native capabilities to work cohesively. It's not a replacement for Claude Code — it's a thin coordination layer that makes Claude Code's built-in features work together autonomously.
Core Philosophy
- Native-First — Use Claude Code's built-in tools, don't reinvent
- Autonomous by Default — Only pause for ambiguity or risk
- Profile-Aware — Adapts to project type (Python, TypeScript, Shell, Go, etc.)
- Quality-Enforced — Blocking gates before commits
- Checkpoint-Safe — Verified commits enable easy rollback
Quick Start
/cs-loop "implement user authentication" # Autonomous work loop
/cs-plan "refactor the API layer" # Plan before executing
/cs-status # Show tasks, git state, profile
/cs-validate # Verify configuration
/cs-learn decision "Use PostgreSQL" "JSON" # Save to memory
/cs-mcp --fix # Register MCP servers
/cs-review 42 # Review a pull request
/cs-init # Create nested CLAUDE.md architecture
/cs-ui # UI/UX audit (web projects)
/cs-team "refactor auth across packages" # Parallel Agent Teams
/cs-docs "feature name" # Generate feature documentation
/cs-assess # Full codebase health audit
/cs-deploy # Deployment readiness check
/cs-sessions # Browse and resume previous sessions
/cs-multi --show # Configure multi-model routing
/cs-debug "TypeError in auth.js" # Structured debugging loop
/cs-log --fix # Logging audit and fix
/cs-batch "migrate all API endpoints to v2" # Parallel worktree fan-out
/cs-schedule "0 2 * * *" /cs-assess # Nightly codebase auditThe Loop
When you invoke /cs-loop, Claude Sentient orchestrates:
1. INIT → Detect profile, load context, fetch library docs
2. UNDERSTAND → Classify complexity (simple/moderate/complex)
3. PLAN → Create tasks (TaskCreate), evaluate team eligibility
4. EXECUTE → Work through tasks (solo or team mode)
5. VERIFY → Run quality gates (lint, test, build)
6. COMMIT → Create checkpoint commit
7. EVALUATE → Done? Exit. More work? Loop.Commands
| Command | Purpose |
/cs-loop [task] | Autonomous work loop |
/cs-plan [task] | Plan before executing |
/cs-status | Show tasks, git state, profile |
/cs-validate | Validate configuration |
/cs-learn [type] [title] [content] | Save to memory (file + MCP) |
/cs-mcp [--test] [--fix] | Check, register, validate MCP servers |
/cs-review [PR] | Review a pull request |
/cs-assess [dir] [--ultrathink] | Full codebase health audit |
/cs-init [dir] | Create/optimize nested CLAUDE.md architecture |
/cs-ui [dir] [--full] | UI/UX audit for web projects |
/cs-team [task] [--status] [--stop] | Create/manage Agent Teams |
/cs-docs ["feature"] [--audit] | Generate and manage feature documentation |
/cs-deploy [--ci] [--docker] [--env] | Deployment readiness check |
/cs-sessions [--search] [--resume] | Browse and resume previous sessions |
/cs-multi [--show] [--set] [--reset] | Configure multi-model orchestration |
/cs-debug [issue] | Structured debug loop: reproduce → trace → isolate → fix → test |
/cs-log [dir] [--fix] [--scaffold] | Logging audit: gaps, unstructured, secrets, correlation IDs |
/cs-batch [plan] | Fan out tasks to parallel worktree agents |
/cs-schedule <cron> <cmd> | Create recurring autonomous agents |
Detailed command structure and skill chaining:
.claude/commands/CLAUDE.md
Project Profiles
| Profile | Detected By | Tools |
| Python | pyproject.toml, *.py | ruff, pytest, pyright |
| TypeScript | tsconfig.json, *.ts | eslint, vitest, tsc |
| Go | go.mod | golangci-lint, go test |
| Rust | Cargo.toml | clippy, cargo test |
| Java | pom.xml, build.gradle | checkstyle, JUnit |
| C/C++ | CMakeLists.txt, Makefile | clang-tidy, ctest |
| Ruby | Gemfile | rubocop, rspec |
| Shell | *.sh, *.ps1 | shellcheck |
| General | (fallback) | auto-detect |
Profile details, model routing, gate structure:
.claude-sentient/profiles/CLAUDE.md
Quality Gates
Before committing, these must pass:
| Gate | Requirement | Auto-Fix |
| LINT | Zero errors from linter | fix_command from profile (up to 3 attempts) |
| TEST | All tests pass | Read test + source, fix logic (never modify assertions) |
| BUILD | Project builds successfully | Read error, fix compilation |
| GIT | Clean working state | Stage changes |
Advisory (report only): TYPE, DOCS, SECURITY.
Gate definitions live in each
.claude-sentient/profiles/*.yamlfile. See.claude-sentient/profiles/CLAUDE.mdfor structure.
Hard Rules
- Use native tools — TaskCreate not custom queue, EnterPlanMode not custom planning
- Never skip gates — Fix issues, don't bypass
- Create checkpoints — Before risky changes
- Respect profiles — Match existing code style
- Ask when ambiguous — One question, then proceed
Integrity rules:
.claude/rules/anthropic-patterns.md| Self-improvement:.claude/rules/learnings.md
Reference
| Topic | Location |
| Memory system | /cs-learn command, .claude/rules/learnings.md |
| Governance files | .claude-sentient/STATUS.md, .claude-sentient/CHANGELOG.md, .claude-sentient/DECISIONS.md |
| Native tools used | TaskCreate, EnterPlanMode, Task, .claude/hooks/*.cjs, MCP servers |
| Native tips | .claude-sentient/documentation/16-native-tips.md |
See @.claude/rules/_index.md for keyword-to-rule mappings and the full rules catalog.
Nested Context Architecture
Detailed documentation lives in nested CLAUDE.md files that load only when needed:
| Location | Content | When Loaded |
.claude/hooks/CLAUDE.md | All 15 hooks, configuration, security patterns, state files | Editing hooks |
.claude-sentient/profiles/CLAUDE.md | Profile detection, model routing, gate structure, conventions | Editing profiles |
.claude/commands/CLAUDE.md | Command structure, documentation policy, rule auto-loading | Editing commands |
Claude Sentient: Orchestrating Claude Code's native capabilities for autonomous development
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

