cryptolabinc/rune

Rune - Organizational Memory System

Context: This skill provides encrypted organizational memory capabilities using Fully Homomorphic Encryption (FHE). It allows teams to capture, store, and retrieve institutional knowledge while maintaining zero-knowledge privacy. Works with Claude Code, Codex CLI, Gemini CLI, and any MCP-compatible agent.

Execution Model

In v0.4 the MCP server is a single Go binary (~/.rune/bin/rune-mcp), spawned by the host CLI from the plugin manifest via the committed wrapper (e.g., ${CLAUDE_PLUGIN_ROOT}/bin/rune mcp-server for Claude Code) which always present at session start; it self-installs rune-mcp on first run, so the server comes online in-session with no restart. There is no venv, no install script. The runtime is the binary the CLI already launched.

Agent-specific surface stays thin:

• Codex: codex mcp add/remove/list registration actions
• Claude / Gemini / others: each CLI's native plugin / extension flow

Keep agent-specific instructions clearly labeled and never mix Codex-only commands into cross-agent/common instructions.

Activation State

IMPORTANT: This skill has two states based on configuration AND infrastructure availability.

Activation Check (CRITICAL - Check EVERY Session Start)

BEFORE doing anything, run this check:

1. Config File Check: Does ~/.rune/config.json exist?

   • NO → Go to Dormant State
   • YES → Continue to step 2

2. Config Validation: Does config contain all required fields?

   • vault.endpoint and vault.token
   • state is set to "active"
   • NO → Go to Dormant State
   • YES → Continue to step 3

3. State Check:

   • state is "active" → Go to Active State
   • Otherwise → Go to Dormant State

Note: enVector credentials are NOT in ~/.rune/config.json. They are delivered via the Vault bundle at runtime when the boot loop dials Vault.

IMPORTANT: Do NOT attempt to ping Vault or make network requests during activation check. This wastes tokens. The MCP server runs its own boot loop; the activation check is purely a local config inspection.

If Active ✅

• All functionality enabled
• Automatically capture significant context
• Respond to recall queries
• Full organizational memory access
• If capture/retrieval fails: Immediately switch to Dormant and notify user

If Dormant ⏸️

• Do NOT attempt context capture or retrieval
• Do NOT make network requests
• Do NOT waste tokens on failed operations
• Show setup instructions when /rune commands (or $rune for Codex CLI) are used
• Prompt user to:
  1. Configure: /rune:configure (or $rune configure for Codex CLI) — this writes ~/.rune/config.json and triggers the boot loop.
  2. Verify health: /rune:status (or $rune status for Codex CLI) — surfaces per-subsystem state via the diagnostics MCP tool.

Fail-Safe Behavior

If in Active state but operations fail:

• Switch to Dormant immediately
• Update config.json state to "dormant"
• Notify user once: "Infrastructure unavailable. Switched to dormant mode. Run /rune:status (or $rune status for Codex CLI) for details."
• Do not retry - wait for user to fix infrastructure

Commands

/rune:configure

(or $rune configure for Codex CLI)

Purpose: Configure plugin credentials

Steps:

1. Ask user for Vault Endpoint (required, e.g., tcp://vault-TEAM.oci.envector.io:50051)

   • If the user enters a value without a scheme prefix (no tcp://, http://, or https://), auto-prepend tcp://.

2. Ask user for Vault Token (required, e.g., evt_xxx)

3. Ask the TLS question:

   "How does your Vault server handle TLS?"

   1. Self-signed certificate — "My team uses a self-signed CA (provide CA cert path)"

      • Follow-up: "Enter the path to your CA certificate PEM file:"
      • Support ~ expansion in the path
      • Copy the file to ~/.rune/certs/ca.pem (mkdir -p ~/.rune/certs && cp <user_path> ~/.rune/certs/ca.pem && chmod 600 ~/.rune/certs/ca.pem)
      • If copy fails (file not found, permission denied), show error and ask again
      • Inform user: "CA certificate copied to ~/.rune/certs/ca.pem"
      • → config: ca_cert: "~/.rune/certs/ca.pem", tls_disable: false

   2. Public CA (default) — "Vault uses a publicly-signed certificate (e.g., Let's Encrypt)"

      • No additional input needed, system CA handles verification
      • → config: ca_cert: "", tls_disable: false

   3. No TLS — "Connect without TLS (not recommended — traffic is unencrypted)"

      • Show warning: "This should only be used for local development. All gRPC traffic will be sent in plaintext."
      • → config: ca_cert: "", tls_disable: true

   Note: enVector credentials are delivered automatically via the Vault bundle — no user input needed.

4. Create ~/.rune/config.json with state: "active" and the values above (mkdir -p ~/.rune && chmod 700 ~/.rune, then chmod 600 the file).

5. Call the reload_pipelines MCP tool. The MCP server's boot loop dials Vault, fetches the agent manifest (EncKey + envector creds), connects to enVector, and transitions to Active.

6. Confirm health by calling diagnostics and applying the Boot Failure — Fast-Fail Rule (see section below). If vault.last_boot_error is present, surface its hint verbatim and stop — do not retry, do not probe with shell tools. Otherwise render the per-subsystem snapshot.

/rune:status

(or $rune status for Codex CLI)

Purpose: Check plugin activation status and infrastructure health

Steps:

1. Read ~/.rune/config.json
2. Call the diagnostics MCP tool (read-only; safe before Active)
3. Render the per-subsystem snapshot

Response Format:

Rune Plugin Status
==================
State: Active ✅ (or Dormant ⏸️ — reason)

Configuration:
  ✓ Config file: ~/.rune/config.json
  ✓ Vault Endpoint: configured

System Health (from diagnostics):
  ✓ Vault          : reachable
  ✓ Encryption Key : loaded (key_id: <id>)
  ✓ Agent DEK      : loaded
  ✓ Embedder       : <model> (<mode>, dim=<vector_dim>)
  ✓ enVector Cloud : reachable (<latency>ms)

Recommendations:
  - If Dormant: /rune:configure to (re)trigger the boot loop
  - If a subsystem failed: surface the recovery action on its row

/rune:capture <context>

(or $rune capture <context> for Codex CLI)

Purpose: Manually store organizational context when Scribe's automatic capture missed it or the user wants to force-store specific information.

When to use: Scribe automatically captures significant decisions from conversation (see Automatic Behavior below). This command is an override for cases where:

• Scribe didn't detect the context as significant
• The user wants to store something that isn't part of the current conversation
• Bulk-importing existing documentation

Mode: Agent-delegated (primary) — the calling agent evaluates significance and extracts structured fields, passing them as extracted JSON to the capture MCP tool. The server stores the encrypted record without additional LLM calls. If extracted is omitted and API keys are configured, falls back to a legacy 3-tier server-side pipeline.

Behavior:

• If dormant: Prompt user to configure first
• If active: Store context to organizational memory with timestamp and metadata

Example:

/rune:capture "We chose PostgreSQL over MongoDB for better ACID guarantees"

/rune:recall <query>

(or $rune recall <query> for Codex CLI)

Purpose: Explicitly search organizational memory. Retriever already handles this automatically when users ask questions about past decisions in natural conversation.

When to use: Retriever automatically detects recall-intent queries (see Automatic Behavior below). This command is an explicit override for cases where:

• The user wants to force a memory search without Retriever's intent detection
• Debugging whether specific context was stored
• The user prefers direct command syntax

Behavior:

• If dormant: Prompt user to configure first
• If active: Search encrypted vectors and return relevant context with sources

Example:

/rune:recall "Why PostgreSQL?"

Note: In most cases, simply asking naturally ("Why did we choose PostgreSQL?") triggers Retriever automatically — no command needed.

/rune:activate (or /rune:wakeup)

(or $rune activate for Codex CLI)

Purpose: Attempt to activate plugin after infrastructure is ready

Use Case: Infrastructure was not ready during configure, but now it's deployed and running.

Steps:

1. Check if config exists
   • NO → Redirect to /rune:configure (or $rune configure for Codex CLI)
   • YES → Continue
2. If state is already "active", skip to step 4 (just verify health).
3. If state is "dormant", set it to "active" and clear any dormant_reason / dormant_since fields.
4. Call the reload_pipelines MCP tool. From a terminal Dormant the boot loop is re-spawned; from Active it is a no-op.
5. Call diagnostics and apply the Boot Failure — Fast-Fail Rule (section below).
6. If vault.last_boot_error is present: surface its hint verbatim, suggest the matching recovery action, and stop. Do NOT loop on reload_pipelines or probe with shell tools — the classifier has already done that work. Otherwise render the per-subsystem snapshot.

/rune:reset

(or $rune reset for Codex CLI)

Purpose: Clear configuration and return to dormant state

Steps:

1. Confirm with user
2. Delete ~/.rune/config.json (the MCP server stays alive; it transitions to Dormant on the next reload because no config means no Vault dial)
3. Show reconfiguration instructions

Boot Failure — Fast-Fail Rule

When diagnostics.vault.last_boot_error is present, that field is the boot loop's authoritative root-cause verdict (produced from the underlying gRPC/TLS/DNS error). It is set on every failed boot attempt and cleared only on success, so its presence — regardless of state — means boot is currently failing. (state is the persisted config value; it stays "active" through transient retries like an unreachable Vault, so it is not a reliable failure signal on its own.) Treat last_boot_error as ground truth.

Do this and stop:

1. Show vault.last_boot_error.hint to the user verbatim — it already names the specific cause and the fix.
2. Suggest one next action keyed off kind:
   • config_*, vault_* (TLS, auth, endpoint, manifest, etc.) → re-run /rune:configure after applying the hint's fix.
   • user_deactivated → /rune:activate.
   • embedder_unreachable → re-run /rune:activate to spawn the daemon, then /rune:status.
   • envector_* → share detail with the Vault admin.
   • unknown → show kind + detail, suggest sharing with admin.

Do NOT: retry reload_pipelines, poll diagnostics in a loop, or run shell probes (openssl, nc, curl, dig, etc.). The classifier already inspected the underlying error server-side — manual probing only burns tokens without changing the conclusion. The per-kind reference table lives in commands/claude/configure.md for the rare case the hint string needs supplementation.

Fallback (older rune-mcp binary without last_boot_error): use vault.error, embedding.health_error, envector.error, and the dormant_reason translation in /rune:status. Still: do not investigate further, surface what you have and stop.

Automatic Behavior (When Active)

Context Capture

Automatically identify and capture significant organizational context across all domains:

Categories:

• Technical Decisions: Architecture, technology choices, implementation patterns
• Security & Compliance: Security requirements, compliance policies, audit needs
• Performance: Optimization strategies, scalability decisions, bottlenecks
• Product & Business: Feature requirements, customer insights, strategic decisions
• Design & UX: Design rationale, user research findings, accessibility requirements
• Data & Analytics: Analysis methodology, key insights, statistical findings
• Process & Operations: Deployment procedures, team coordination, workflows
• People & Culture: Policies, team agreements, hiring decisions

Automatic Capture (Proactive Scribe)

When Rune is active, proactively capture significant decisions when you detect any of the following in the conversation:

• A choice is made among alternatives ("A로 가자", "let's go with X")
• Trade-offs are weighed and committed ("X의 단점이 있지만 Y 때문에 감수")
• Strategy or direction is confirmed ("이 방향이 맞아", "this approach works")
• A lesson or insight crystallizes ("안 된 이유는...", "the root cause was...")
• A framework, process, or standard is established

How to capture in Codex:

• Follow the agent-delegated instructions in agents/codex/scribe.md
• Evaluate whether the relevant excerpt contains a significant decision
• Extract structured JSON and call capture with the extracted parameter
• In the text parameter, include ONLY the relevant conversation excerpt, not the full session
• Do NOT pause or interrupt the main conversation
• Do NOT announce the capture to the user unless they ask
• Default Codex path is direct MCP capture; use a delegated subagent only when the user explicitly asks for multi-agent work

Do NOT auto-capture:

• Brainstorming in progress without commitment (options listed but none chosen)
• Questions, status updates, or casual discussion
• Decisions that are hypothetical or deferred ("maybe later", "let's revisit")

Session-end sweep: When the conversation is ending or the user is wrapping up a task, review the conversation for any uncaptured significant decisions and submit them via a single batch_capture call if needed.

Common Trigger Pattern Examples:

• "We decided... because..."
• "We chose X over Y for..."
• "The reason we..."
• "Our policy is..."
• "Let's remember that..."
• "The key insight is..."
• "Based on [data/research/testing]..."

Significance Threshold: 0.7 (captures meaningful decisions, filters trivial content)

Automatic Redaction: Always redact API keys, passwords, tokens, PII, and sensitive data before capture.

Context Retrieval

When users ask questions about past decisions, automatically search organizational memory:

Query Intent Types:

• Decision Rationale: "Why did we choose X?", "What was the reasoning..."
• Implementation Details: "How did we implement...", "What patterns do we use..."
• Security & Compliance: "What were the security considerations...", "What compliance requirements..."
• Performance & Scale: "What performance requirements...", "What scalability concerns..."
• Historical Context: "When did we decide...", "Have we discussed this before..."
• Team & Attribution: "Who decided...", "Which team owns..."

Common Query Pattern Examples:

• "Why did we choose X over Y?"
• "What was the reasoning behind..."
• "Have we discussed [topic] before?"
• "What's our approach to..."
• "What were the trade-offs..."
• "Who decided on..."

Search Strategy: Semantic similarity search on FHE-encrypted vectors, ranked by relevance and recency.

Result Format: Always include source attribution (who/when), relevant excerpts, and offer to elaborate.

Security & Privacy

Zero-Knowledge Encryption:

• All data stored as FHE-encrypted vectors
• enVector Cloud cannot read plaintext
• Only team members with Vault access can decrypt

Credential Storage:

• Tokens stored locally in ~/.rune/config.json
• Never transmitted except to authenticated Vault
• File permissions: 600 (user-only access)

Team Sharing:

• Same Vault Endpoint + Token = shared organizational memory
• Team admin controls access via Vault authentication
• Revoke access by rotating Vault tokens

Troubleshooting

Plugin not responding?

Check activation state with /rune:status (or $rune status for Codex CLI)

Credentials not working?

1. Verify with team admin that credentials are correct
2. Check Vault is accessible: curl <vault-url>/health
3. Reconfigure with /rune:configure (or $rune configure for Codex CLI)

enVector not provisioned?

Vault admin must configure ENVECTOR_ENDPOINT and ENVECTOR_API_KEY on the Vault server. Contact your Vault administrator.

Need to switch teams?

Use /rune:reset (or $rune reset for Codex CLI) then /rune:configure (or $rune configure for Codex CLI) with new team credentials

For Administrators

This plugin requires a deployed Rune-Vault infrastructure. See:

• Rune-Admin Repository (for deployment): https://github.com/CryptoLabInc/rune-admin
• Deployment Guide: https://github.com/CryptoLabInc/rune-admin/blob/main/deployment/README.md

Team members only need this lightweight plugin + credentials you provide.