Meta-Router
Automatically indexes and routes user intents to installed OpenClaw skills by scanning SKILL.md files and dynamically injecting relevant skill contexts.
MIT-0 · Free to use, modify, and redistribute. No attribution required.
⭐ 1 · 30 · 1 current installs · 1 all-time installs
MIT-0
Security Scan
OpenClaw
Suspicious
medium confidencePurpose & Capability
The skill's name/description (meta-routing, lazy-loading skills) align with the runtime steps (listing ~/.openclaw/skills, reading SKILL.md, building an index and loading specific SKILL.md). However several declared constraints are odd: requiring the index to remain under 2KB is impractical for many installed skills and the SKILL.md refers to writing '.meta_index.json in the root directory' (ambiguous — could mean repository root, home, or system root), which is not justified by the purpose.
Instruction Scope
Runtime instructions direct the agent to 'intercept user input before processing' and 'perform silent background scanning' and to read arbitrary SKILL.md files via shell commands (ls, cat). Intercepting all user input and performing silent scans is a privacy/overreach concern: it gives the skill visibility into every prompt and may change control flow without explicit consent. Dynamically mounting arbitrary SKILL.md into the model context is consistent with routing but also broad — it will cause the agent to load other skills' instructions on demand, which could unintentionally surface malicious or sensitive content present in those files.
Install Mechanism
Instruction-only skill with no install spec or code files — lowest install risk. All file interactions happen via runtime shell commands described in SKILL.md rather than bundled installers or downloads.
Credentials
No credentials or env vars are requested (good), but the skill expects to read and write files under the user's profile (e.g., ~/.openclaw/skills/* and a .meta_index.json written to an unclear 'root'). Access to those file paths is consistent with indexing installed skills, but writing an index to an ambiguous location and performing silent scans triggered by keywords is disproportionate to a minimally intrusive indexer.
Persistence & Privilege
always: false (normal). The skill's AGENT.md instructs it to check the skills directory at start of every session and to monitor prompts for keywords (effectively persistent monitoring within agent sessions). This is not a platform-level privilege escalation, but it does mean the skill will run routinely and can intercept user input unless explicitly disabled.
What to consider before installing
This skill does what it claims (indexing and routing skills) but has several ambiguous or surprising behaviors you should consider before installing: it asks the agent to silently intercept user input and automatically scan your ~/.openclaw/skills directory, it will write a .meta_index.json file to an unclear location (check where that ends up), and it enforces an unrealistic 2KB size limit for the index which could cause truncation or repeated scans. Recommended precautions: (1) inspect a few installed SKILL.md files to ensure they contain no secrets or harmful instructions; (2) ask the maintainer where exactly .meta_index.json will be written and whether it could overwrite existing files; (3) prefer explicit/manual refresh triggers rather than silent monitoring if you want tighter control; and (4) if you rely on sensitive or private skills, test Meta-Router in a confined environment first. If any of these answers are unsatisfactory, treat the skill as high-risk and avoid installing it.Like a lobster shell, security has layers — review code before you run it.
Current versionv1.0.0
Download zipAutomationEfficiencyInfrastructurePerformanceRoutingUtilitylatest
License
MIT-0
Free to use, modify, and redistribute. No attribution required.
SKILL.md
Skill: Meta-Router-Automata
Description
Automates the management of the OpenClaw skill library. Implements self-scanning and real-time indexing via shell commands to eliminate token bloat in multi-skill environments.
Automation Procedures
1. Self-Indexing (The Trigger)
- Activation Logic:
- Triggered automatically after a new skill installation.
- Triggered by explicit user request ("refresh library").
- Triggered if
.meta_index.jsonis missing.
- Workflow:
- Execute
ls -d ~/.openclaw/skills/*/to map all installed skill paths. - Traverse paths and extract the first two lines (ID and Description) of every
SKILL.md. - Aggregate the results (Skill Name, Path, and Brief).
- Generate or overwrite
.meta_index.jsonin the root directory.
- Execute
- Output Schema:
{"[Shortcut]": {"id": "[Folder_Name]", "desc": "[Summary]"}}
2. Intent Dispatching (Routing)
- Execution:
- Intercept user input before processing.
- Perform fuzzy matching against keywords in
.meta_index.json. - Priority: [Explicit Shortcut] > [Semantic Intent Match] > [General Mode].
- Dynamic Mounting: Once a skill is identified, use
cator system commands to read itsSKILL.mdand inject it into the current context.
3. Context Pruning
- After routing is successful, explicitly discard irrelevant skill cached metadata to maximize context window (90%+ focus on the current task).
Constraints
- The index file must remain under 2KB.
- Scanning must be performed silently in the background without interrupting the user flow.
Files
3 totalSelect a file
Select a file to preview.
Comments
Loading comments…
