Features

Everything you need to capture, validate, and transfer AI agent expertise.

Built for How You Work

Developers: npm/pip for agent expertise, with a type checker

Scientists: A validated, composable lab notebook for AI agents

ML Engineers: Model registry, but for agent knowledge instead of weights

Teams: Institutional knowledge capture for AI workflows

🎯

Graduated Compiler

A 7-pass validation pipeline that catches errors early and auto-fixes common issues. Three severity tiers: hard errors block writes, auto-corrections fix silently, advice suggests improvements.

  • Pass 1-2: Syntax (ast.parse) and structure (find class)
  • Pass 3-4: Type inference and schema check
  • Pass 5: Type check with auto-corrections (W001-W003)
  • Pass 6-7: Execute in sandbox and semantic validation
  • 10 error codes (E001-E010), 6 auto-corrections (W001-W006), 5 advice codes (I001-I006)
validation output

$ stato validate .stato/


PASS qc.py (skill)

W001 depends_on is string, auto-wrapping in list (auto-fixed)

W003 Version missing patch number: '1.0' → '1.0.0' (auto-fixed)

I003 No lessons_learned on skill


PASS plan.py (plan)


All 2 module(s) valid.

🔄

Session Resume

Restore agent context after /compact or at the start of a new session. Reads all modules and produces a structured recap.

  • Full recap: project context, plan progress, skills with parameters, memory state
  • Brief mode: --brief for a one-paragraph summary
  • Raw output: --raw for plain text (pipeable)
  • Works after /compact: modules survive on disk, resume reads them fresh
context restoration

$ stato resume --brief


cortex_scrna: scRNA-seq analysis of mouse cortex P14. Progress: 3/7 steps complete. Next: find_hvg. QC and normalization complete.


$ stato resume

+----- Project Resume ---------------+

| Project: cortex_scrna |

| Plan: 3/7 steps complete |

| Next: Step 4, find_hvg |

| Skills: qc_filtering v1.2.0 |

| normalize v1.1.0 |

| Phase: analysis |

+------------------------------------+

🧩

Composition Algebra

Four operations for managing portable expertise archives.

  • Snapshot: Export all or selected modules as a .stato archive. Template mode resets progress, preserves lessons.
  • Slice: Extract specific modules with dependency tracking. --with-deps auto-includes transitive dependencies.
  • Graft: Add external modules with conflict detection. Choose: replace, rename, skip, or ask.
  • Import: Restore from archive with optional filtering by module or type.
composition

$ stato snapshot --name "expert" --template

✓ Created expert.stato (4 modules, 3.8 KB)


$ stato slice --module skills/qc --with-deps

Auto-included: skills/normalize.py (dependency)

✓ Created slice.stato


$ stato import expert.stato --path ~/new-project

+ skills/qc.py

+ skills/normalize.py

+ plan.py

+ memory.py

🌐

Cross-Platform Bridges

Generate lightweight bridge files (~500 tokens) that point agents to your expertise modules. Same .stato/ directory, every platform.

  • Claude Code: CLAUDE.md (auto-read)
  • Cursor: .cursorrules (auto-read)
  • Codex: AGENTS.md (auto-read)
  • Generic: README.stato.md (manual)
  • On-demand loading: bridge is a table of contents, agents load full skill files only when needed
bridge generation

$ stato bridge --platform all

✓ Generated CLAUDE.md

✓ Generated .cursorrules

✓ Generated AGENTS.md

✓ Generated README.stato.md

🌍

Web AI Bundle Import

Transfer expertise from web AI conversations into coding agent projects. One file, one command.

  • Bundle format: single Python file with SKILLS dict + PLAN/MEMORY/CONTEXT strings
  • Safe parsing: uses ast.parse (no exec on untrusted files)
  • Full validation: every module runs through the 7-pass compiler before writing
  • Dry run: --dry-run to preview without writing
bundle import

$ stato import-bundle stato_bundle.py --platform all


Bundle contents:

Skills: 3 (qc, normalize, clustering)

Plan: yes

Memory: yes

Context: yes


Result: 6 imported, 0 failed

✓ Generated CLAUDE.md

✓ Generated .cursorrules

✓ Generated AGENTS.md

✓ Generated README.stato.md

🛡

Privacy Scanner

19 regex patterns across 6 categories detect secrets, PII, and sensitive paths before you share archives.

  • Categories: api_key, credential, token, path, network, pii
  • Interactive review: [s]anitize, [r]eview, [f]orce, [c]ancel
  • Sanitize-on-export: replaces secrets with placeholders, originals unchanged
  • .statoignore: suppress false positives with pattern matching
privacy scan

$ stato snapshot --name "share"


Privacy scan found 3 item(s):


api_key (1 found)

context.py:8 — API key (OpenAI)

sk-abc123... → {API_KEY}


path (2 found)

context.py:12 — Home directory path

/home/niki/... → /home/{user}/...


Choose action [s/r/f/c]: s

✓ Secrets sanitized in archive