SlipBox Skill

IMPORTANT: Before doing anything else, run the setup check below. Do not skip this step.

Setup Check

Run this command immediately upon skill invocation:

echo "SLIPBOX_API_KEY: ${SLIPBOX_API_KEY:+${SLIPBOX_API_KEY:0:6}…(set)}" | sed 's/^SLIPBOX_API_KEY: $/SLIPBOX_API_KEY: (MISSING)/'
echo "SLIPBOX_URL: ${SLIPBOX_URL:-(MISSING)}"
echo "SLIPBOX_PRIVATEBOX_REPO: ${SLIPBOX_PRIVATEBOX_REPO:-(MISSING)}"

If any show (MISSING):

• STOP IMMEDIATELY. Do not attempt any further action.
• Do not guess values, use defaults, search for shell config files, or attempt fallbacks of any kind.
• Tell the user exactly which variables are missing and that they must set them in ~/.zshrc (or ~/.zprofile) and re-source their shell before trying again.
• End your response there and wait for the user to fix the issue.

Once env vars are confirmed, verify the service is reachable:

curl -sL "$SLIPBOX_URL/api/health"
# {"status":"ok"}

If the health check fails or returns anything other than {"status":"ok"}:

• STOP IMMEDIATELY. Do not attempt any further action.
• Report the response to the user and tell them the service is unavailable.
• End your response there and wait for the user.

API Error Handling

If any API call returns an error response (any JSON with an "error" field, or a non-2xx HTTP status):

• STOP IMMEDIATELY. Do not attempt any further action.
• Do not try to write notes directly to PrivateBox or any other fallback.
• Do not retry with different parameters or modified requests.
• Report the exact error response to the user and wait for them to resolve it.

---

About

Interact with the SlipBox semantic knowledge engine and browse your PrivateBox notes.

SlipBox service: $SLIPBOX_URL PrivateBox repo: $SLIPBOX_PRIVATEBOX_REPO

Configuration

Required environment variables (set in shell):

SLIPBOX_API_KEY=<shared-secret>          # Bearer token for API auth
SLIPBOX_URL=https://slip-box-rho.vercel.app  # SlipBox service base URL
SLIPBOX_PRIVATEBOX_REPO=Randroids-Dojo/PrivateBox   # GitHub repo for notes (owner/repo)

All other configuration (OpenAI, GitHub, PrivateBox) lives on the deployed Vercel service.

---

Quick Reference

All API calls require: Authorization: Bearer $SLIPBOX_API_KEY

# Health check
curl -sL "$SLIPBOX_URL/api/health"

# Add a note
curl -sL -X POST "$SLIPBOX_URL/api/add-note" \
  -H "Authorization: Bearer $SLIPBOX_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"content": "Atomic idea goes here."}'

# Add a typed note (type: "meta" or "hypothesis")
curl -sL -X POST "$SLIPBOX_URL/api/add-note" \
  -H "Authorization: Bearer $SLIPBOX_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"content": "## Cluster: ...", "type": "meta"}'

# Re-link all notes (recompute similarity links)
curl -sL -X POST "$SLIPBOX_URL/api/link-pass" \
  -H "Authorization: Bearer $SLIPBOX_API_KEY"

# Cluster notes into thematic groups
curl -sL -X POST "$SLIPBOX_URL/api/cluster-pass" \
  -H "Authorization: Bearer $SLIPBOX_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"k": 5}'

# Detect conceptual tensions (contradictions within clusters)
curl -sL -X POST "$SLIPBOX_URL/api/tension-pass" \
  -H "Authorization: Bearer $SLIPBOX_API_KEY"

# Fetch theme data (clusters + note content + tensions) for agent synthesis
curl -sL "$SLIPBOX_URL/api/theme-data" \
  -H "Authorization: Bearer $SLIPBOX_API_KEY"

---

API Reference

POST /api/add-note

Capture an atomic idea. SlipBox embeds it, links it to similar notes, and commits it to PrivateBox.

Optional type field sets a semantic type in the note's frontmatter. Valid values: "meta" (AI-generated cluster summary) or "hypothesis" (AI-generated research hypothesis). Omit for regular atomic notes.

# Regular note
curl -sL -X POST "$SLIPBOX_URL/api/add-note" \
  -H "Authorization: Bearer $SLIPBOX_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "content": "The Zettelkasten method treats each note as a discrete, reusable idea."
  }'

# Meta-note (cluster synthesis)
curl -sL -X POST "$SLIPBOX_URL/api/add-note" \
  -H "Authorization: Bearer $SLIPBOX_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "content": "## Cluster: Agentic Systems\n\nNotes in this cluster explore...",
    "type": "meta"
  }'

Response:

{
  "noteId": "20260222T153045-a1b2c3d4",
  "type": "meta",
  "linkedNotes": [
    {"noteId": "20260110T091200-b2c3d4e5", "similarity": 0.91},
    {"noteId": "20260115T143000-c3d4e5f6", "similarity": 0.85}
  ]
}

type is null in the response for regular notes.

POST /api/link-pass

Recompute semantic similarity links across all notes. Run after adding many notes in bulk.

curl -sL -X POST "$SLIPBOX_URL/api/link-pass" \
  -H "Authorization: Bearer $SLIPBOX_API_KEY"

Response:

{"message": "Link pass complete", "notesProcessed": 42, "totalLinks": 156}

POST /api/cluster-pass

Run k-means clustering on note embeddings. Omit k to auto-select cluster count.

curl -sL -X POST "$SLIPBOX_URL/api/cluster-pass" \
  -H "Authorization: Bearer $SLIPBOX_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"k": 5}'

Response:

{
  "message": "Cluster pass complete",
  "noteCount": 42,
  "clusterCount": 5,
  "clusters": [{"id": "cluster-0", "size": 9, "noteIds": ["20260222T153045-a1b2c3d4", ...]}, ...]
}

POST /api/tension-pass

Detect conceptual tensions — notes with contradictory content that cluster near each other.

curl -sL -X POST "$SLIPBOX_URL/api/tension-pass" \
  -H "Authorization: Bearer $SLIPBOX_API_KEY"

Response:

{
  "message": "Tension pass complete",
  "noteCount": 42,
  "clusterCount": 5,
  "tensionCount": 8,
  "tensions": [{"id": "tension-0", "noteA": "...", "noteB": "...", "similarity": 0.68, "clusterId": "cluster-0"}, ...]
}

GET /api/theme-data

Returns clusters with full note content and tensions for local LLM agent synthesis. No embeddings — only human-readable data. Use this to read your knowledge graph and synthesize meta-notes per cluster, then POST them back via /api/add-note.

Requires a current clusters index (run cluster-pass first).

curl -sL "$SLIPBOX_URL/api/theme-data" \
  -H "Authorization: Bearer $SLIPBOX_API_KEY"

Response:

{
  "clusters": [
    {
      "id": "cluster-0",
      "noteIds": ["20260222T153045-a1b2c3d4", "20260110T091200-b2c3d4e5"],
      "notes": {
        "20260222T153045-a1b2c3d4": {"title": "Optional title", "body": "Atomic idea content."},
        "20260110T091200-b2c3d4e5": {"body": "Another idea."}
      }
    }
  ],
  "tensions": [
    {"id": "tension-0", "noteA": "20260222T153045-a1b2c3d4", "noteB": "20260110T091200-b2c3d4e5", "similarity": 0.65, "clusterId": "cluster-0"}
  ],
  "clusterCount": 1,
  "noteCount": 2,
  "tensionCount": 1,
  "computedAt": "2026-02-23T01:33:00.000Z"
}

If no clusters exist yet, returns { "message": "No clusters found. Run cluster-pass first.", "clusters": [], ... }.

---

Note Format

Notes in PrivateBox are Markdown files with YAML frontmatter:

---
id: 20260222T153045-a1b2c3d4
title: "Optional title"
type: meta
tags: ["tag1", "tag2"]
source: "URL or origin"
created: 2026-02-22T15:30:45.000Z
updated: 2026-02-22T15:30:45.000Z
links:
  - target: 20260110T091200-b2c3d4e5
    similarity: 0.91
  - target: 20260115T143000-c3d4e5f6
    similarity: 0.85
---

Atomic idea content in Markdown.

type is omitted for regular notes. Valid values: meta, hypothesis.

Note ID format: YYYYMMDDTHHMMSS-<8hex> (timestamp + content hash)

Index files (in index/ directory of PrivateBox):

• index/embeddings.json — noteId → vector + model + timestamp
• index/backlinks.json — noteId → array of linking notes
• index/clusters.json — thematic groups of notes
• index/tensions.json — pairs of contradictory notes

---

Reading Notes from PrivateBox

Use the gh CLI to read notes without needing direct GitHub API calls.

# List all notes
gh api "repos/$SLIPBOX_PRIVATEBOX_REPO/contents/notes" \
  --jq '.[].name'

# Read a specific note by ID
gh api "repos/$SLIPBOX_PRIVATEBOX_REPO/contents/notes/20260222T153045-a1b2c3d4.md" \
  --jq '.content' | base64 -d

# Read the backlinks index
gh api "repos/$SLIPBOX_PRIVATEBOX_REPO/contents/index/backlinks.json" \
  --jq '.content' | base64 -d | jq '.'

# Read the clusters index
gh api "repos/$SLIPBOX_PRIVATEBOX_REPO/contents/index/clusters.json" \
  --jq '.content' | base64 -d | jq '.'

# Read the tensions index
gh api "repos/$SLIPBOX_PRIVATEBOX_REPO/contents/index/tensions.json" \
  --jq '.content' | base64 -d | jq '.'

Search notes by content

# Search PrivateBox notes for a keyword (uses GitHub code search)
gh api "search/code?q=<keyword>+repo:$SLIPBOX_PRIVATEBOX_REPO+path:notes" \
  --jq '.items[].path'

# Or clone locally for fast full-text search
gh repo clone "$SLIPBOX_PRIVATEBOX_REPO" /tmp/privatebox
grep -r "keyword" /tmp/privatebox/notes/ --include="*.md" -l

Find notes by tag

gh api "search/code?q=tags+keyword+repo:$SLIPBOX_PRIVATEBOX_REPO+path:notes" \
  --jq '.items[].path'

---

Workflows

Capture an idea

1. Write the atomic idea as a single focused thought.
2. POST to /api/add-note with the content.
3. Note the returned noteId and linkedNotes to see what concepts it connects to.
4. Optionally run /api/link-pass afterward if adding many notes in bulk.

Browse and explore

1. List notes with gh api "repos/$SLIPBOX_PRIVATEBOX_REPO/contents/notes".
2. Read individual notes by ID.
3. Follow links in frontmatter to explore connected ideas.
4. Use the clusters index to browse thematic groups.
5. Use the tensions index to identify areas of conceptual conflict worth investigating.

Synthesize theme meta-notes

After running a full analysis cycle, use GET /api/theme-data to read every cluster with its note contents, then write one atomic synthesis note per cluster and POST each back via /api/add-note.

# 1. Fetch theme data
curl -sL "$SLIPBOX_URL/api/theme-data" \
  -H "Authorization: Bearer $SLIPBOX_API_KEY"

# 2. For each cluster, synthesize a meta-note and POST it back with type "meta"
curl -sL -X POST "$SLIPBOX_URL/api/add-note" \
  -H "Authorization: Bearer $SLIPBOX_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "content": "## Cluster: Agentic Systems\n\nNotes in this cluster explore...",
    "type": "meta"
  }'

The type: meta field is written to the note's frontmatter and can be used by future passes to filter AI-generated notes from atomic notes.

Run a full analysis cycle

After adding a batch of notes or to refresh the graph:

# Step 1: Recompute links
curl -sL -X POST "$SLIPBOX_URL/api/link-pass" \
  -H "Authorization: Bearer $SLIPBOX_API_KEY"

# Step 2: Recluster
curl -sL -X POST "$SLIPBOX_URL/api/cluster-pass" \
  -H "Authorization: Bearer $SLIPBOX_API_KEY"

# Step 3: Detect tensions
curl -sL -X POST "$SLIPBOX_URL/api/tension-pass" \
  -H "Authorization: Bearer $SLIPBOX_API_KEY"