Auto Invoke Router

SKILL.md

---
name: auto-invoke-router
version: 1.0.0
description: Scans installed skills and AGENTS.md to generate a routing config that maps conversation intents to skills. Use when you want to reduce manual skill invocation, improve intent matching across a large skill library, or when setting up OpenClaw for the first time with multiple skills installed.
author: ordo-tech
tags: [routing, skills, intent, automation, agents, config, discovery]
metadata:
  openclaw:
    requires:
      bins:
        - find
        - bash
    emoji: "🔀"
    homepage: https://clawhub.com/@ordo-tech/auto-invoke-router
---

## What this skill does

Reads every installed skill's `SKILL.md` frontmatter (the `description` field) and any local `AGENTS.md` to produce a `skill_routing` config block. This config maps intent keywords and phrases to skill names, giving OpenClaw a structured reference for selecting the right skill based on conversation context.

The output is written into `AGENTS.md` as a `## Skill Routing` section, or saved as a standalone `router.yml`. Native auto-invoke behaviour depends on your OpenClaw version — check your release notes or docs to confirm whether `skill_routing` in `AGENTS.md` is read automatically. In all versions, the config serves as a clear, human-readable routing reference that can be wired up manually or extended with custom rules.

## When to use it

- More than ~5 skills are installed and manual invocation is getting unwieldy
- Users are triggering the wrong skill or missing relevant ones
- Setting up a new OpenClaw instance and want smart defaults from the start
- After installing a batch of skills and wanting to refresh the routing map
- Whenever the `available_skills` block in context feels cluttered or mismatched

## Usage

### Step 1 — Locate installed skills

Derive the OpenClaw system skill directory from the binary location:
```bash
OPENCLAW_BIN=$(which openclaw 2>/dev/null || which claw 2>/dev/null)
OPENCLAW_SYSTEM_SKILLS=$(dirname "$OPENCLAW_BIN")/../lib/node_modules/openclaw/skills
find ~/.openclaw/skills "$OPENCLAW_SYSTEM_SKILLS" -name "SKILL.md" 2>/dev/null
```

If `which openclaw` returns nothing, also try common install locations:
```bash
find ~/.openclaw/skills \
  /opt/homebrew/lib/node_modules/openclaw/skills \
  /usr/local/lib/node_modules/openclaw/skills \
  -name "SKILL.md" 2>/dev/null
```

Collect the full list of `SKILL.md` paths. If `clawhub list` is available, also run:
```bash
clawhub list
```
to confirm installed skill names.

### Step 2 — Extract descriptions

For each `SKILL.md` found, read the YAML frontmatter block (lines between the opening and closing `---`). Extract:
- `name` — the skill identifier
- `description` — the full triggering description

Do not read the body of each SKILL.md; frontmatter only.

If a skill is missing a `name` or `description` field, skip it and note it in the final report as: `skill-x: skipped — missing description`. Do not fabricate a description.

### Step 3 — Read AGENTS.md

Read `AGENTS.md` in the current workspace. Look for it at `./AGENTS.md` relative to the workspace root, or at `~/.openclaw/workspace/AGENTS.md` if no workspace context is set. Identify any existing `## Skill Routing` section. If it exists, it will be fully replaced in Step 6.

### Step 4 — Generate intent keywords

For each skill, derive 3–8 intent keywords or short phrases from the description. Rules:
- Use lowercase
- Prefer noun phrases and verb phrases that a user would naturally say (e.g. "search for skill", "install skill", "weather forecast", "github PR", "security audit")
- Omit generic words: "use", "when", "skill", "tool", "this"
- Include negatives if the description calls them out explicitly (e.g. "NOT for historical data")

After generating all triggers, check for conflicts: if the same keyword appears under two or more different skills, flag it in the report as ambiguous. Do not remove the keyword — leave it in both entries and let the user resolve it.

### Step 5 — Produce the routing config

Output a YAML block in this format:

```yaml
# auto-invoke-router — generated by auto-invoke-router skill
# Regenerate by invoking: auto-invoke-router
# WARNING: this section is fully regenerated on each run. Manual edits will be
# overwritten. To preserve custom triggers, add them above this block in AGENTS.md
# with a comment like: # custom-routing-preserve
skill_routing:
  version: "1.0"
  rules:
    - skill: clawhub
      triggers:
        - search clawhub
        - install skill
        - update skill
        - publish skill
        - clawhub list
    - skill: weather
      triggers:
        - weather
        - temperature
        - forecast
        - rain
    - skill: gh-issues
      triggers:
        - github issue
        - fix bug
        - open PR
        - pull request
        - review comments
    # ... one entry per installed skill
  fallback: null  # set to any installed skill name to invoke when no rule matches,
                  # or leave as null to take no default action
```

Use the actual installed skill names and generated triggers — the above is illustrative only.

### Step 6 — Write output

**Option A — Append to AGENTS.md (recommended):**

Add a `## Skill Routing` section at the end of `AGENTS.md` containing the full `skill_routing:` YAML block inside a fenced code block. If a `## Skill Routing` section already exists, replace it in full — all triggers are regenerated from current descriptions. Any manual edits to the previous section will be lost; users should preserve custom triggers outside this block (see the warning comment in Step 5).

**Option B — Standalone file:**

Write the YAML block to `router.yml` in the workspace root. Inform the user to reference it in `AGENTS.md` if they want OpenClaw to pick it up automatically.

### Step 7 — Report

After writing, output a short summary:

```
Router updated — N skills mapped, M skipped
Skills covered: skill-a, skill-b, skill-c, ...
Skipped (no description): skill-x, skill-y
Trigger conflicts (review manually): keyword-foo (skill-a, skill-b), keyword-bar (skill-c, skill-d)
Output: AGENTS.md § Skill Routing (or router.yml)
Note: verify your OpenClaw version supports native skill_routing before relying on auto-invoke.
```

## Examples

### Example 1: Fresh install with 6 skills

**Input:** 6 installed skills found via `find`. `AGENTS.md` has no routing section.

**Output:** `## Skill Routing` appended to `AGENTS.md` with 6 rule entries, each containing 4–6 triggers derived from their descriptions. Summary reports "6 skills mapped, 0 skipped."

### Example 2: Refresh after installing new skills

**Input:** `AGENTS.md` already has a `## Skill Routing` section with 4 rules. 3 new skills were installed since last run.

**Output:** Existing section replaced in full with 7 rules — all triggers regenerated from current descriptions. Summary reports "7 skills mapped." Any manual trigger edits in the previous section are not preserved; the report reminds the user to re-apply them if needed.

### Example 3: No AGENTS.md found

**Input:** Workspace has no `AGENTS.md`.

**Output:** `router.yml` written to workspace root. User informed to create `AGENTS.md` and include the routing block, or to re-run once `AGENTS.md` is present.

### Example 4: Skills with missing descriptions

**Input:** 8 skills found; 2 have no `description` field in their frontmatter.

**Output:** 6 rules generated for the skills with descriptions. Report lists the 2 skipped skills by name. User can add descriptions to those skills' `SKILL.md` files and re-run.

## Requirements

- At least one skill installed with a valid `description` field (output is empty otherwise — report this clearly)
- Read access to `~/.openclaw/skills/` and OpenClaw's system skill directories
- Write access to `AGENTS.md` or workspace root for `router.yml`
- `bash` tool available for the `find` and `which` commands in Step 1

No API keys or external accounts required. Runs entirely on local files.

## Support

Issues or questions: https://clawhub.com/@ordo-tech/auto-invoke-router
Publisher: @ordo-tech on ClawHub