agentleFS
SanStone3SanStone3/agent-strata/skills/strata/references/claude.md

agent-strata / references

CLAUDE.md2 starsChanged 39 days agoMITView on GitHub

Read this reference only for Claude Code installation, routing, or troubleshooting. Aliases resolve according to provider and can change. Pin full model IDs only when reproducibility and provider rollout require it. Personal skills and agents live in ~/.claude/skills/ and ~/.claude/agents/; project equivalents live in .claude/skills/ and .claude/agents/.

  • In this repositoryOne other file it gives its agents.

The file

We split this claude.md into the rules it gives an agent

4 rules. About 0 are common, 2 are distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Read this reference only for Claude Code installation, routing, or troubleshooting.

    line 3
  2. Very specific:

    Read-only agents list only Read, Grep, and Glob.

    line 48
  3. Distinctive:

    If the agents directory did not exist when the session started, restart Claude Code after first installation.

    line 49
  4. Very specific:

    Start the highest-tier main session with claude --agent fable-controller.

    line 55

≈ “Common” is an estimate: at least 1 in 5 of the 27,641 CLAUDE.md 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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
claude.md

Claude Code configuration

Read this reference only for Claude Code installation, routing, or troubleshooting.

Recommended roles

AgentModel aliasEffortToolsUse
Default primaryopus[1m]xhighsession permissionsGoal ownership, integration
haiku-scouthaikuinheritedRead, Grep, GlobLocate paths and evidence
sonnet-workersonnetxhighread/write/bashOrdinary bounded implementation
opus-workeropusxhighread/write/bashComplex/high-risk implementation
opus-revieweropusxhighread-only toolsConsequential review
fable-controllerfablexhighsession permissionsExplicit highest-tier main session
fable-workerfablexhighread/write/bashExceptional long implementation
fable-reviewerfablexhighread-only toolsExceptional high-consequence review

Aliases resolve according to provider and can change. Pin full model IDs only when reproducibility and provider rollout require it.

Concurrency and nesting

ControlMechanismDefaultStrata position
Concurrent subagentsCLAUDE_CODE_MAX_CONCURRENT_SUBAGENTS20Respect actual host capacity and explicit user/project budgets; Strata adds no fixed numeric limit
Spawn depthCLAUDE_CODE_MAX_SUBAGENT_SPAWN_DEPTH3 layers below the main conversationSet to 1 so the no-descendant invariant is enforced by the client, not only by instructions
  • Claude Code's cap is far above the default budget, so the budget is a deliberate policy choice about write safety and synthesis capacity rather than a client restriction. Raise it only by the checklist in scaling.md.
  • A spawn past the cap fails with Concurrent subagent limit reached and instructs no retry. Queue the packet into the next wave.
  • Forks occupy a slot while running, and resuming a completed subagent takes a fresh slot without a limit check, so a resume-heavy plan can exceed the intended budget without any client error.
  • Sessions running at the maximum effort level are exempt from the concurrent cap. There the stated budget is the only limit, so state the number before dispatching.
  • Both variables take a positive whole number and belong in the env block of the settings file being merged. CLAUDE_CODE_MAX_SUBAGENT_SPAWN_DEPTH = 1 disables nesting entirely.

Files

  • ../assets/templates/claude/settings-snippet.json
  • ../assets/templates/claude/CLAUDE-snippet.md
  • ../assets/templates/claude/agents/*.md

Personal skills and agents live in ~/.claude/skills/ and ~/.claude/agents/; project equivalents live in .claude/skills/ and .claude/agents/.

Merge notes

  • User settings are ~/.claude/settings.json; shared project settings are .claude/settings.json; personal project overrides are .claude/settings.local.json.
  • Merge JSON keys. Never replace existing permissions, plugins, hooks, env, or organization-defined controls. Add CLAUDE_CODE_MAX_SUBAGENT_SPAWN_DEPTH into the existing env object instead of replacing it, and leave an existing concurrency variable alone unless the user asked for a different ceiling.
  • effortLevel accepts persistent values through xhigh; max is session-only. The templates deliberately use xhigh for the main session and all coding/review agents.
  • Frontmatter effort overrides the session level unless an environment variable or managed control takes precedence.
  • Some third-party providers currently resolve sonnet or opus aliases to 4.6-class models that do not support xhigh; Claude Code lowers the request to the highest supported level. Installation is complete only after /status or equivalent evidence confirms the resolved model and effort. Otherwise pin an organization-approved full model ID that supports xhigh, or report this invariant as unmet.
  • Read-only agents list only Read, Grep, and Glob; workers list editing and Bash tools. Tool lists reduce accidental capability but do not grant permission beyond the parent session.
  • If the agents directory did not exist when the session started, restart Claude Code after first installation. Environment changes in settings also require a new session.

Fable

Fable requires a sufficiently recent Claude Code version and account/organization access. It is explicit rather than default because availability and billing behavior vary.

  • Start the highest-tier main session with claude --agent fable-controller.
  • Delegate fable-worker or fable-reviewer only after the user explicitly selected Fable or availability/approved usage credits are known.
  • Interactive Claude Code can request consent before billing usage credits. Non-interactive print-mode and Agent SDK flows may bill without that prompt; do not select Fable there without prior authorization.
  • Authentication, billing, rate-limit, request-size, and transport errors do not activate model fallback. If Fable is unavailable, stop the branch and return control without retry loops.

Official sources

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.