strata
Skill2 starsChanged 39 days agoMITView on GitHub
Cost-aware native subagent orchestration for Codex or Claude Code. Choose models, task effort, parallelism and compact context for independent coding, research or review; do not use for trivial sequential work.
4 other projects use a version of this.
Adopted by
Most recently taken up by
- momori777/Artemistheir own version, first seen Oct 25, 2026
- AhmedNazihX/Stratatheir own version, first seen Oct 2, 2026
- rx6ru/stratatheir own version, first seen Aug 27, 2026
- alsk1992/strata-agent-skillstheir own version, first seen Jul 30, 2026
Other versions4 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 4
- momori777/Artemis★ 445same name
- AhmedNazihX/Strata★ 4same name
- alsk1992/strata-agent-skills★ 0same name
- rx6ru/strata★ 0same 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.
In this repositoryOne other file it gives its agents.
- skills/strata/references/claude.mdCLAUDE.md
The file
We split this skill into the rules it gives an agent
2 rules. 0 are common, 2 are distinctive, and 0 are very specific to this project.
0
Common
Most of the 3 edited versions we read keep these.
2
Distinctive
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 29
Use the active native tool's verified model/effort list.
- Distinctive: line 33
Read only what the task needs: installation, Codex, Claude, login/SSH failures, validation.
“Common” is measured: kept by at least 6 in 10 of the 3 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Strata
Optimize cost and elapsed time per accepted change, including context, retries and review. One controller owns integration and quality.
Route by task
| Task | Codex starting point |
| Known path, short search, small sequential edit | Direct tools |
| Bounded extraction, search, summary or exact mechanical execution | Luna low/medium; scout read-only or executor |
| Coding and agent work with understood scope | Sol medium; worker |
| Readily checked scoped fix, supplied-material writing, vision or computer use | Terra medium/high; scoped worker |
| Cross-module diagnosis, architecture, security, concurrency or difficult implementation | Sol high/xhigh; deep_worker |
| Consequential independent review | Sol high/xhigh; reviewer, read-only |
These are starting points, not quality guarantees. Read-only work can need Sol. Terra's lower token price does not make Terra xhigh cheaper than Sol medium. Use cost-control.md for dated evidence; retain required tests and evidence.
Orchestrate
- The current CLI/App user selection owns the primary model and effort, including later user switches. Never restore a startup value or write primary settings during skill updates. Sol xhigh is a recommendation only. Children keep their creation model; escalation uses a new compact packet.
- Strata has no numeric agent cap. Choose useful parallelism from actual host capacity, resources, cost and synthesis capacity. Default to one writer; allow more only with disjoint paths, frozen contracts and independent checks. Queue excess work. See scaling.md.
- Use the active native tool's verified model/effort list. Prefer explicit new-child overrides and history-free packets. Resolve missing bindings once through model-routing.md; avoid rediscovery on every task. Automatic choices: reviewed Sol, Terra and Luna; Astra needs an explicit approved pin. No GPT-5.6.
- Send goal, file/contract pointers, owned paths or read-only scope, constraints, checks, stop condition and a concise result budget. Reuse evidence; do not copy transcripts or duplicate searches. Return findings/changes, locations, tests and uncertainty. See task-contract.md when boundaries are unclear.
- Scouts/reviewers are read-only. Children do not spawn descendants, expand scope, commit, push, deploy or invoke another coding-agent provider. Preserve CODEX_HOME, auth, provider and permissions. Inspect changes before integration; inspect partial state before retrying.
Read only what the task needs: installation, Codex, Claude, login/SSH failures, validation.
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.

