NextSAFE

Claude Code plugin that generates individualized knowledge systems from conversation. You describe how you think and work, have a conversation and get a complete second brain as markdown files you own.

The instruction file, verbatim from the audited commit — this is the text the model reads, and the surface the audit's instruction layer examines. Quoted here so you can judge it without cloning anything.

---
name: next
description: Surface the most valuable next action by combining task stack, queue state, inbox pressure, health, and goals. Recommends one specific action with rationale. Triggers on "/next", "what should I do", "what's next".
version: "1.0"
generated_from: "arscontexta-v1.6"
user-invocable: true
context: fork
model: sonnet
allowed-tools: Read, Grep, Glob, Bash
---

## Runtime Configuration (Step 0 — before any processing)

Read these files to configure domain-specific behavior:

1. **`ops/derivation-manifest.md`** — vocabulary mapping, domain context
   - Use `vocabulary.notes` for the notes folder name
   - Use `vocabulary.inbox` for the inbox folder name
   - Use `vocabulary.note` for the note type name in output
   - Use `vocabulary.topic_map` for MOC references
   - Use `vocabulary.cmd_reduce` for process/extract command
   - Use `vocabulary.cmd_reflect` for connection-finding command
   - Use `vocabulary.cmd_reweave` for backward-pass command
   - Use `vocabulary.rethink` for rethink command name

2. **`ops/config.yaml`** — thresholds, processing preferences
   - `self_evolution.observation_threshold` (default: 10)
   - `self_evolution.tension_threshold` (default: 5)

If these files don't exist, use universal defaults and generic command names.

## EXECUTE NOW

**INVARIANT: /next recommends, it does not execute.** Present one recommendation with rationale. The user decides what to do. This prevents cognitive outsourcing where the system makes all work decisions and the user becomes a rubber stamp.

**Execute these steps IN ORDER:**

---

### Step 1: Read Vocabulary

Read `ops/derivation-manifest.md` (or fall back to `ops/derivation.md`) for domain vocabulary mapping. All output must use domain-native terms. If neither file exists, use universal terms (notes, inbox, topic map, etc).

---

### Step 2: Reconcile Maintenance Queue

Before collecting state, evaluate all maintenance conditions and reconcile the queue. This ensures maintenance tasks are current before the recommendation engine runs.

**Read queue file** (`ops/queue/queue.json` or `ops/queue.yaml`). If `schema_version` < 3, migrate:
- Add `maintenance_conditions` section with default thresholds
- Add `priority` field to existing tasks (default: "pipeline")
- Set `schema_version: 3`

**For each condition in maintenance_conditions:**

1. **Evaluate the condition:**

| Condition | Evaluation Method |
|-----------|------------------|
| orphan_notes | For each note in {vocabulary.notes}/, count incoming [[links]]. Zero = orphan. |
| dangling_links | Extract all [[links]], verify targets exist as files. Missing = dangling. |
| inbox_pressure | Count *.md in {vocabulary.inbox}/. |
| observation_accumulation | Count status: pending in ops/observations/. |
| tension_accumulation | Count status: pending or open in ops/tensions/. |
| pipeline_stalled | Queue tasks with status: pending unchanged across sessions. |
| unprocessed_sessions | Count files in ops/sessions/ without mined: true. |
| moc_oversize | For each topic map, count linked notes. |
| stale_notes | Notes not modified in 30+ days with < 2 links. |
| low_link_density | Average link count across all notes. |
| methodology_drift | Compare config.yaml modification time vs newest ops/methodology/ note modification time. If config is newer, methodology may be stale. |

2. **If condition exceeds threshold AND no pending task with this condition_key exists:**

Create maintenance task:
```bash
TIMESTAMP=$(date -u +"%Y-%m-%dT%H:%M:%SZ")
MAINT_MAX=$(jq '[.tasks[] | select(.id | startswith("maint-")) | .id | ltrimstr("maint-") | tonumber] | max // 0' ops/queue/queue.json)
NEXT_MAINT=$((MAINT_MAX + 1))

jq --arg id "maint-$(printf '%03d' $NEXT_MAINT)" \
   --arg priority "{priority}" \
   --arg key "{condition_key}" \
   --arg target "{description}" \
   --arg action "{recommended command}" \
   --arg ts "$TIMESTAMP" \
   '.tasks += [{"id": $id, "type": "maintenance", "priority": $priority, "status": "pending", "condition_key": $key, "target": $target, "action": $action, "auto_generated": true, "created": $ts}]' \
   ops/queue/queue.json > tmp.json && mv tmp.json ops/queue/queue.json
```

3. **If condition is satisfied AND a pending task with this condition_key exists:**

Auto-close it:
```bash
TIMESTAMP=$(date -u +"%Y-%m-%dT%H:%M:%SZ")
jq --arg key "{condition_key}" --arg ts "$TIMESTAMP" \
   '(.tasks[] | select(.condition_key == $key and .status == "pending")).status = "done" |
    (.tasks[] | select(.condition_key == $key and .status == "pending")).completed = $ts' \
    ops/queue/queue.json > tmp.json && mv tmp.json ops/queue/queue.json
```

4. **If condition fires AND a pending task already exists:**

Update the target description (specifics may have changed):
```bash
jq --arg key "{condition_key}" --arg target "{new description}" \
   '(.tasks[] | select(.condition_key == $key and .status == "pending")).target = $target' \
   ops/queue/queue.json > tmp.json && mv tmp.json ops/queue/queue.json
```

---

### Step 3: Collect Vault State

Gather all signals. Run independent checks in parallel where possible. Record each signal even if the check returns zero — absence of signal is itself informative.

| Signal | How to Check | What to Record |
|--------|--------------|----------------|
| **Task stack** | Read `ops/tasks.md` — current priorities and open items | Top items, open count, any deadlines |
| **Queue state** | Read `ops/queue.yaml` or `ops/queue/queue.json` — pending pipeline tasks | Total pending, by phase (create, reflect, reweave, verify), blocked phases |
| **Inbox pressure** | Count `*.md` files in {vocabulary.inbox}/, find oldest by mtime | Count per subdirectory, age of oldest item in days |
| **Note count** | Count `*.md` in {vocabulary.notes}/ | Total notes for context |
| **Orphan notes** | For each note, grep for `[[filename]]` across all files — zero hits = orphan | Count, first 5 names |
| **Dangling links** | Extract all `[[links]]` from notes/,

SAFEgrade B · trust 89/100 Nothing in the source contradicts what it says it does. Grade A is reserved for packages that have also passed the behavioural sandbox.

What the source does

Filesystem

none-observed

Network

none-observed

Shell

none-observed

Dependencies

pinned

Secrets in source

none-found

Findings (0)

No findings outside the package's declared scope.

Gates applied: no_behavioural_pass.