- Home
- Skills
- Williamzujkowski
- Cognitive Toolworks
- Core Gemini Delegator
core-gemini-delegator_skill
- Python
5
GitHub Stars
3
Bundled Files
2 months ago
Catalog Refreshed
4 months ago
First Indexed
Readme & install
Copy the install command, review bundled files from the catalogue, and read any extended description pulled from the listing source.
Installation
Preview and clipboard use veilstrat where the catalogue uses aiagentskills.
npx veilstrat add skill williamzujkowski/cognitive-toolworks --skill core-gemini-delegator- CHANGELOG.md3.3 KB
- index-entry.json1.0 KB
- SKILL.md6.5 KB
Overview
This skill automates fast delegation decisions to the Gemini CLI (via MCP) for tasks involving large files or deep analysis. It enforces a deterministic routing rule: delegate when a single file exceeds 100KB or when the task is analysis/review/summarization or sandbox code testing. The skill returns a compact decision payload with rationale and any security flags.
How this skill works
The skill validates input schema, scans provided file paths for sensitive patterns, and checks file sizes. It then applies a two-step Tier‑1 decision: a delegation gate (thresholds and task-type checks) and an output formatter that emits either a Gemini MCP delegation_command or a recommendation to keep the task in Claude. It enforces a T1 token budget for fast, repeatable decisions and includes security aborts when secrets are detected.
When to use it
- Single file >100KB needing comprehension, summarization, or code review
- Tasks explicitly labeled analysis, review, summarize, or codebase-review
- Sandboxed execution of unfamiliar code for safe testing
- When you need deterministic, fast routing based on file size and task type
- When you want a short, actionable delegation_command for Gemini MCP
Best practices
- Always validate task_description is non-empty before calling the decision routine
- Pre-scan file paths for .env, credentials.json, *.key, *.pem and abort on matches
- Include a concise context_size_estimate when available to improve determinism
- Keep multi-file orchestration out of this skill; use a dedicated orchestration agent
- Ensure delegation_command contains no API keys, secrets, or PII before execution
Example use cases
- Summarize a 3.5MB research paper: delegate to gemini-mcp with a short analyze-and-summarize command
- Review a large codebase file (>100KB): delegate to Gemini for comprehension and highlighted issues
- Run safe sandbox tests of an unfamiliar script: delegate to Gemini for isolated execution and results
- Refactor a 45KB function: keep in Claude for fine-grained edits and precise suggestions
- Decide routing programmatically in a CI step based on file size and task metadata
FAQ
Any single file over 100KB or tasks tagged analysis/review/summarize/codebase-review trigger delegation to Gemini.
What happens if secrets are found in file paths?
The skill aborts delegation, adds security_flags describing the findings, and asks the user to sanitize files before retry.
Is this skill suitable for multi-file orchestration?
No. Complex multi-file orchestration should be handled by a separate orchestration agent.