ContextSutra AI
Build a hashed project capsule from docs, decisions, and manifests.
Local-first CLI that turns documentation, decisions, tasks, and manifest metadata into a verifiable capsule. npm [email protected] published 2026-11-10. The README says it does not read code logic and does not call a model.
- Offline
- 12 skills
- Node 20+
- npm latest
- [email protected]
- GitHub version
- 0.2.1
- Licence
- Apache-2.0
- Language
- HTML
- Weekly downloads
- 529
- Overall downloads
- 529
- GitHub stars
- 4
Live npm weekly downloads, npm overall downloads, and the latest npm version, plus GitHub stars and the package.json version, fetched 2026-11-17 19:07 UTC. Cached 10 minutes. Weekly downloads cover 2026-11-10 to 2026-11-16. Overall downloads cover 2026-10-01 to 2026-11-17. npm does not publish an earlier total. Stars and downloads show public use. They are not a quality score.
Use it when
- 1
You want a generated brief an assistant can read, with a staleness check.
- 2
You want secret-looking lines redacted before anyone shares the capsule.
- 3
You want one capsule rendered for Claude Code, Cursor, Codex, Gemini CLI, or Copilot.
What it does, and what it does not
Read this before the guide. A limit here is a limit of the tool.
Does
- scan builds a bounded, read-only capsule of docs, ADRs, tasks, and manifest metadata.
- verify lists which documents are stale or new. The README says the integrity digest is a checksum, not a digital signature.
- review is a privacy gate: BLOCK or HUMAN_REVIEW_REQUIRED. The README says it never says safe.
- Decisions count as approved only if the source says Status: accepted.
- Instruction-like lines are quarantined and never rendered, including a line that tells a model to ignore rules.
- 12 skills. The README says host activation is untested. It records a 47-check self-audit and 54 tests.
Does not
- It does not read code logic. It captures documentation, decisions, tasks, and manifest metadata.
- Redaction is regex best effort. The README says it misses secrets that do not match known patterns and does not detect personal data.
- It does not call a model and has no network code, as stated in the README. The self-audit is in-process, not an independent attestation.
- Copying skills into a host folder does not prove the IDE loaded them.
- 0 runtime dependencies. Node.js 20 or newer.
Claims above are from the project pages and are attributed as stated there. They are not independently verified here.
How a run moves
From the ContextSutra README, npm 0.2.0, published 2026-11-10. The capsule is generated. It is not a hand-written AGENTS.md.
-
1. Input
Docs, decisions, tasks
Plus manifest metadata. Code logic is not read. Secret files are not read.
-
2. Capsule
Hashed and line-referenced
The digest is a checksum, not a signature. Instruction-like lines are quarantined.
-
3. Review
Privacy gate
BLOCK or HUMAN_REVIEW_REQUIRED. The README says the gate never says safe.
-
4. Freshness
verify
Lists stale or new documents. Exit 2 means the capsule is stale.
How to read the result
Wording is from the project page or README and is attributed as stated there. It is not independently verified here.
Commands and notes
-
npx [email protected] demo --out contextsutra-demosample capsule, hand-off, and report. - contextsutra scan --root . --out .contextsutra/capsule.json
- contextsutra review --capsule .contextsutra/capsule.json
-
contextsutra verify --root . --capsule .contextsutra/capsule.jsonthe README says exit 2 if stale. -
contextsutra install --root . --host claudealso cursor, codex, gemini, copilot, generic.
Guide
How to use the result. GitHub is the source repository. npm is the install.
What a capsule is
From the ContextSutra README, npm 0.2.0, published 2026-11-10. Local. Node.js 20 or newer.
- scan reads docs, ADRs, tasks, and manifest metadata. It does not read code logic. Secret files are not read.
- The integrity digest is a checksum, not a digital signature. Decisions count as approved only if the source says Status: accepted.
- Instruction-like lines are quarantined and never rendered, including a line that tells a model to ignore rules.
- review returns BLOCK or HUMAN_REVIEW_REQUIRED. The README says it never says safe. verify exits 2 if the capsule is stale.
What is untested
- Redaction is regex best effort. The README says it misses secrets that do not match known patterns and does not detect personal data.
- 12 skills. Host activation is untested. The self-audit is in-process, not an independent attestation. The README records 54 tests and a 47-check self-audit.
- It does not call a model and has no network code. 0 runtime dependencies.
- Add .contextsutra/ to .gitignore if the capsule may hold internal text.
Documentation
Titles below are the HTML titles from the GitHub Pages site on 2026-11-17. A sentence is shown only when that page has its own meta description. This website did not re-check the measurements those pages describe.
- ContextSutra AI Local-first, keyless CLI for verifiable, privacy-aware project context capsules and 12 portable agent skills.
- ContextSutra AI documentation Complete documentation for ContextSutra AI: executive summary, functional spec, architecture, technical reference, user guide, security, legal, testing evidence and audit.
- 01. Executive summary · ContextSutra AI 01. Executive summary: ContextSutra AI documentation by Vatsal Shah.
- 02. Functional specification · ContextSutra AI 02. Functional specification: ContextSutra AI documentation by Vatsal Shah.
- 03. Architecture · ContextSutra AI 03. Architecture: ContextSutra AI documentation by Vatsal Shah.
- 04. Technical reference · ContextSutra AI 04. Technical reference: ContextSutra AI documentation by Vatsal Shah.
- 05. User guide · ContextSutra AI 05. User guide: ContextSutra AI documentation by Vatsal Shah.
- 06. Skills catalog · ContextSutra AI 06. Skills catalog: ContextSutra AI documentation by Vatsal Shah.
- 07. Security and legal · ContextSutra AI 07. Security and legal: ContextSutra AI documentation by Vatsal Shah.
- 08. Testing and evidence · ContextSutra AI 08. Testing and evidence: ContextSutra AI documentation by Vatsal Shah.
- 09. Audit and trust · ContextSutra AI 09. Audit and trust: ContextSutra AI documentation by Vatsal Shah.
- 10. CI/CD and release · ContextSutra AI 10. CI/CD and release: ContextSutra AI documentation by Vatsal Shah.
- 11. FAQ and glossary · ContextSutra AI 11. FAQ and glossary: ContextSutra AI documentation by Vatsal Shah.
- 12. Contributor handbook · ContextSutra AI 12. Contributor handbook: ContextSutra AI documentation by Vatsal Shah.
- 13. Roadmap and status · ContextSutra AI 13. Roadmap and status: ContextSutra AI documentation by Vatsal Shah.
- 14. Architecture decisions · ContextSutra AI 14. Architecture decisions: ContextSutra AI documentation by Vatsal Shah.
Questions
Does ContextSutra send the repo to a model?
No. The README says it never calls a model and has no network code.
Is a capsule the same as CLAUDE.md?
No. The README says those files are hand-written. A capsule is generated, hashed, line-referenced, and checkable for staleness.
Other projects
- VigilSutra AI 256 last week · 2,190 overall · 4 stars
- RachnaSutra AI 208 last week · 925 overall · 4 stars
- ArchSutra AI 176 last week · 883 overall · 4 stars
- PramanaSutra AI 195 last week · 750 overall · 4 stars
- EvalSutra AI 207 last week · 563 overall · 4 stars
- DocSutra AI 511 last week · 511 overall · 4 stars
- IncidentSutra AI 490 last week · 490 overall · 4 stars
- MigraSutra AI 519 last week · 519 overall · 4 stars
- WattSutra AI 440 last week · 440 overall · 4 stars
- ProtocolSutra AI 530 last week · 530 overall · 4 stars
- EdgeSutra AI 337 last week · 337 overall · 4 stars
- ContribSutra AI 274 last week · 274 overall · 4 stars
- LocalSutra AI 295 last week · 295 overall · 4 stars
- LineageSutra AI 0 last week · 0 overall · 0 stars
- DeploySutra AI 0 last week · 0 overall · 0 stars