DocNexus — Local Project Memory for Agents
Local, source-attributed project memory for coding agents: capture reviewed Markdown, search and recall it offline, and keep architecture decisions grounded in evidence.
Free
Works with the AI Agents you already use
DocNexus — Local Project Memory for Agents
Example session with this skill installed
Use DocNexus to answer from our project memory:
Why do refresh tokens rotate on every use, where is that decision recorded, and what evidence supports it? Cite the stored source and call out any conflicts.
- Read your context and instructions
- Compiled the docnexus
Answer
DocNexus found one current decision record and one supporting source. The policy is to rotate a refresh token on every successful use, issue a replacement pair, and revoke the complete token family if an already-used token is replayed.
Evidence
| Claim | Evidence | Location |
|---|---|---|
| Refresh tokens rotate after every successful refresh | Threat-model decision and implementation note | docnexus/library/auth/token-rotation.md |
| Reuse revokes the full token family | Security review accepted 2026-10-14 | docnexus/records/auth-001/metadata.json |
| Revocation is atomic with token issuance | Transaction boundary noted in implementation | src/auth/token-service.ts |
| Two older records conflict about rotation timing | Legacy note says daily rotation; replacement record supersedes it | docnexus/records/auth-000/timeline.md |
Current decision
The active policy rotates the refresh token on every use. The new token is written in the same transaction that invalidates the previous token. Detection of reuse revokes the whole family rather than returning a generic authentication error.
Provenance
- Source:
docnexus/records/auth-001/source.md - Timeline:
docnexus/records/auth-001/timeline.md - Current document:
docnexus/library/auth/token-rotation.md - Superseding record:
auth-001 - Older conflicting record:
auth-000
Next steps
- Treat
auth-001as the authoritative decision. - Run
docnexus document get --id auth-001 --include document,source,timelinebefore changing token behavior. - Update the affected knowledge page if the transaction boundary changes.
Connects securely to your tools. The creator never sees your data.
About this skill
DocNexus — Local Project Memory for Agents
DocNexus keeps curated project knowledge in Markdown next to a source-attributed timeline. Six coordinated skills let Codex, Claude Code, and other SKILL.md agents capture decisions, retrieve prior context, cite evidence, maintain indexes, and recover from Git changes without sending project knowledge to an external LLM.
What it does
- Durable project memory — current Markdown documents plus records containing the original source, metadata, timeline, and provenance.
- Hybrid recall — keyword retrieval reads text directly; semantic recall fuses keyword and local vector ranks with one-hop graph evidence.
- Offline by default — the bundled quantized ONNX model runs locally; the CLI makes no external LLM calls.
- Review before write — capture flows extract and seal drafts, show a preview, and only ingest after the user authorizes the change.
- Maintenance — validates canonical text, rebuilds PGLite or Postgres/pgvector projections, and repairs derived concept and graph indexes.
- Agent-native — the six skills route capture, extraction, ingestion, recall, library operations, and maintenance automatically.
How it works
Install the npm package inside the target project, initialize an agent integration, then ask the agent to remember a decision or answer from memory. A write runs through draft → seal → preview → authorization → ingest. Reads return complete documents and library paths for citation instead of unverified snippets.
Requirements and breaking change
Requires Node.js 22.13+ and npm. Workspace format 7 is used by current releases. On versions 0.24.2+, npm postinstall deletes recognized legacy workspaces from formats 1–6 without confirmation, migration, or backup. Back up any existing workspace before installing. Source checkouts and unrecognized directories are not affected.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for LLM Ops.
Current workspace format 7 requires fresh initialization. Installing 0.24.2+ deletes recognized legacy workspaces (formats 1–6) without migration or backup, so back up first. Refined documents are capped at 4,000 Unicode characters. Semantic recall requires a local index; keyword reads still work without it. Git hooks skip non-Git projects, and rebases or renames may need manual document-level recovery.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Free to use
No account required to browse
Trust & safety
Security scan
Security scan passed 5 days ago
- Free with an Agensi account
Linux only
- Software
Node.js
Gitnpmcachesvitestinstallscache
- Optional
- PostgreSQLpgvector
Install into the target project using the README command. Back up existing DocNexus data before upgrading from an older workspace format. Writes stay gated until you review and approve the sealed draft.
Permissions
File Scopes
Read/write access covers the .docnexus workspace, the configured Markdown knowledge directory, Git hooks, and agent instruction files. Network and environment variables are used only for npm dependencies or optional external PostgreSQL; the CLI makes no external LLM calls.