context-router
Skill0 starsChanged 50 days agoMITView on GitHub
Retrieve the smallest sufficient repository context for an agent task, with deterministic local search and optional explicit Jev shadow scoring. Use for context selection, context-cost reduction, or evidence routing; do not use it to bypass authoritative instructions.
- Reads credentials
1 other project uses a version of this.
Adopted by
Most recently taken up by
- llopresto87/Cypresstheir own version, first seen Oct 14, 2026
Other versions1 project uses a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 1
- llopresto87/Cypress★ 119same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
The file
We split this skill into the rules it gives an agent
1 rule. About 0 are common, 1 is distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 10
Run local discovery first:
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Context Router
Use this skill to discover task-relevant repository evidence without giving workers unrelated source text. Start with a small required kernel: the user task, repository instructions, current changes, and the relevant runtime/build facts. Use the local router to identify candidate facets, then open only the evidence needed to resolve the task.
Run local discovery first:
python3 <skill-path>/scripts/context_router.py index --project . --out <private-index.jsonl>
python3 <skill-path>/scripts/context_router.py search --index <private-index.jsonl> --query "<task summary>"
python3 <skill-path>/scripts/context_router.py task-search --index <private-index.jsonl> --query "<task summary>"task-search groups related task/spec/plan/evidence records by task ID without dumping their full text. Inspect prior-task details only when they may carry a still-binding decision, failed approach, migration constraint, regression evidence, or unfinished acceptance criterion. Do not load history merely because vocabulary overlaps; current instructions, accepted requirements, and live code remain authoritative.
packet assembles a local worker-packet skeleton from the search results: task, role, owned paths, an acceptance placeholder for the coordinator, scored facets, grouped prior-task facts and a provenance note. start scaffolds tasks/<task-id>/{spec,plan,evidence}.md plus architecture/README.md, reports existing prior-task records, and prints guidance to partition still-binding history into the new artifacts. check compares the index against the current tree and reports added/removed/changed paths so a stale index is re-built before it misleads.
Use results to build a worker packet with task, owned files, acceptance criteria, applicable constraints, selected prior-task facts, and direct source pointers. Record which prior task and artifact supplied each historical constraint. Never use a result to omit repository instructions, an accepted requirement, or live evidence needed to resolve a conflict.
Remote Jev scoring is optional shadow telemetry only. Read the remote policy before using it. It requires both deliberate privacy approval and --allow-remote; a key alone is not approval. The key may be supplied through a project .env (TYPESAFE_API_KEY), and CONTEXT_ROUTER_JEV=0 hard-disables remote scoring. On any failure, use the normal context process without blocking the task. The per-task sequence, output interpretation and fallback rules are in the Jev usage runbook.
The installer is preview-first and additive for Codex, Claude Code, OpenCode and Kimi Code. It copies the skill into each runner's skill directory, adds a /context-route command for Claude Code and OpenCode, and appends a small managed activation block to AGENTS.md (Codex, OpenCode, Kimi Code) or CLAUDE.md (Claude Code). The block is loaded at session start, so local index/search becomes the default for substantive new sessions without requiring /context-route. No activation authorizes remote source transmission. See the init guide for setup, first use, reinstall and removal steps.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.