fusebase-devfusebase-dev/fusebase-flow/AGENTS.md

fusebase-flow

AGENTS.md9 starsChanged 27 days agoMITView on GitHub

This repository uses Fusebase Flow for lifecycle discipline and Fusebase CLI provider assets for runtime/domain guidance. The recovery-owned overlay below is the portable bootstrap for every agentic surface. Canonical rules live in FLOW_RULES.md; do not duplicate them here. Fusebase Flow owns the development lifecycle; existing Fusebase CLI runtime, MCP, SDK, provider, and project rules remain authoritative for their domains. Project-specific rules win on conflict. Full boundary map: docs/fusebase-cli-edition.md. A file is read again only if its full text is not…

Where files are headingNew AGENTS.md files have more rules about keeping secrets out of the code, and fewer about code style.
Of the AGENTS.md files we read, the 31,843 first seen in the last six months, against the older ones: the kind of rule that gained most share, and the kind that lost most. About files like this in general, not this one, and an estimate, since a kind is a pattern in the words.
In this repository119 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- .github/copilot-instructions.mdCopilot instructions
- .cursor/rules/fusebase-flow-always.mdcCursor rule
- .cursor/rules/fusebase-flow-implementation.mdcCursor rule
- .cursor/rules/fusebase-flow-security.mdcCursor rule
- .cursor/rules/fusebase-flow-specs.mdcCursor rule
- .cursor/rules/fusebase-flow-validation.mdcCursor rule
- api-explorationSkill
- app-api-contract-testingSkill
- app-backendSkill
- app-business-docsSkill
- app-dev-practicesSkill
- app-e2e-testsSkill
- app-quality-patternsSkill
- app-routingSkill
- app-secretsSkill
- app-sidecarSkill
- app-ui-designSkill
- business-logic-guardianSkill
- client-vs-internalSkill
- code-reviewSkill
- comment-policySkill
- communicationSkill
- design-discovery-ideationSkill
- dev-debug-logsSkill
- documentation-budgetSkill
- file-uploadSkill
- find-wasted-codeSkill
- find-wasted-effortSkill
- fusebase-cliSkill
- fusebase-dashboardsSkill
- fusebase-flow-health-checkSkill
- fusebase-gateSkill
- fusebase-portal-specific-appsSkill
- git-history-diagnosticSkill
- git-workflowSkill
- handling-authentication-errorsSkill
- handoffSkill
- implementation-planningSkill
- invite-with-passwordSkill
- lightweight-laneSkill
- liveness-disciplineSkill
- managed-integrationsSkill
- mcp-gate-debugSkill
- module-size-disciplineSkill
- north-starSkill
- phase-auditSkill
- product-apps-decompositionSkill
- product-docs-firstSkill
- product-ownerSkill
- project-onboardingSkill
- release-deploy-reportingSkill
- remote-logsSkill
- repo-onboarding-context-mapSkill
- requirements-specificationSkill
- role-disciplineSkill
- security-permissions-reviewSkill
- skill-authoringSkill
- smoke-testingSkill
- task-delegationSkill
- The first 60 shown.

The file

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

4 rules. About 0 are common, 1 is distinctive, and 3 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.
3
Very specific
Tied to their project's names, files or tools. Probably skip.

Common
Distinctive
Very specific

Very specific:
Read FLOW_RULES.md through ## Amendment log.
line 15
Very specific:
Read flow-skills/communication/SKILL.md and flow-skills/role-discipline/SKILL.md once unless each exact body is already in context.
line 16
Very specific:
Read the active workflow and ticket/handoff artifacts.
line 18
Distinctive:
Follow FR-01 through FR-27 and the attested role's don't-list.
line 33

≈ “Common” is an estimate: at least 1 in 5 of the 40,246 AGENTS.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

AGENTS.md

AGENTS.md — Fusebase Flow portable bootstrap

FuseBase Flow — workflow lifecycle overlay

Fusebase Flow owns the development lifecycle; existing Fusebase CLI runtime, MCP, SDK, provider, and project rules remain authoritative for their domains. Project-specific rules win on conflict. Full boundary map: docs/fusebase-cli-edition.md.

Session bootstrap

Read FLOW_RULES.md through ## Amendment log; it is the single resident safety/ownership core.
Read flow-skills/communication/SKILL.md and flow-skills/role-discipline/SKILL.md once unless each exact body is already in context. A skill name or description does not count as its body.
Self-attest one role using FLOW_RULES.md, then read flow-skills/role-discipline/references/<role>.md.
Read the active workflow and ticket/handoff artifacts. If docs/north-star.md, an applicable product doc, or a business-logic index/doc exists, read it; do not create absent onboarding artifacts.

Surface	Startup delivery	Required action
Claude Code	skill descriptions/metadata, not bodies	read both mandatory bodies + role reference
Codex	optional skill descriptions/metadata, not bodies	read both mandatory bodies + role reference
Cursor	always-on adapter text, not skill bodies	read both mandatory bodies + role reference
Copilot / VS Code	repository instruction adapters, not skill bodies	read both mandatory bodies + role reference
Gemini-style IDEs	no Flow skill bodies	read both mandatory bodies + role reference
Delegated sub-agent	inherits no mandatory body	delegating prompt inlines the digest + push blocks (`task-delegation`); a code-edit worker also reads `FLOW_RULES.md` + `references/ai-developer.md`, a read-only worker only what its brief names

A file is read again only if its full text is not already in context (token-economy TE-02); a host may inject only part of it (Codex loads the first 32 KiB of AGENTS.md by default), so check for the FuseBase Flow — workflow lifecycle overlay section, not for the injection. Provider adapters point here and to canonical owners; they do not reprint protocol bodies. On-demand skills live in flow-skills/; workflows in workflows/; policies in policies/; hooks in hooks/; templates in templates/. Most portable commands map to their same-named skill; where no such skill exists (/step-out), read the canonical command body. Canonical command bodies and provider command adapters both live under hooks/local/fusebase-flow-overlays/commands/.

Safety and Git

Follow FR-01 through FR-27 and the attested role's don't-list. Operator questions and approvals stay in chat text.
One T-task per commit. Stage exact paths; never git add ., git add -A, --no-verify, force-push, hard-reset, clean, or recursive-delete without the explicit authorization required by FLOW_RULES.md and policy.
Save cross-session handoffs before chat output. Stop at the verification gate; deploy requires the owning role and authorization contract.
Runtime/domain rules from CLI provider skills override generic implementation guidance. Flow still owns specs, decisions, tasks, gates, reviews, deploy handoffs, and smoke contracts.

Portable commands

Command	Claude Code	Codex (`/prompts:<cmd>` if installed)	Portable (any agent)
`/product-owner`	`/product-owner`	`/prompts:product-owner`	invoke the `product-owner` skill
`/onboard`	`/onboard`	`/prompts:onboard`	invoke the `project-onboarding` skill
`/handoff`	`/handoff`	`/prompts:handoff`	invoke the `handoff` skill
`/fusebase-health`	`/fusebase-health`	`/prompts:fusebase-health`	invoke the `fusebase-flow-health-check` skill
`/token-waste-audit`	`/token-waste-audit`	`/prompts:token-waste-audit`	invoke the `token-economy` skill
`/find-wasted-effort`	`/find-wasted-effort`	`/prompts:find-wasted-effort`	invoke the `find-wasted-effort` skill
`/find-wasted-code`	`/find-wasted-code`	`/prompts:find-wasted-code`	invoke the `find-wasted-code` skill
`/step-out`	`/step-out`	`/prompts:step-out`	read `hooks/local/fusebase-flow-overlays/commands/step-out.md`

Installation and update safety

Merge existing consumer content; never overwrite AGENTS.md, CLAUDE.md, .gitignore, .claude/settings.json, .mcp.json, .cursor/mcp.json, fusebase.json, .codex-plugin/plugin.json, skills-lock.json, existing .agents/skills/, .claude/skills/, or .github/workflows/. Procedure: docs/install-fusebase-cli-project.md.

Flow ≤4.6.1 → only bash hooks/local/bootstrap-upgrade.sh -- --auto-yes; Flow 4.7.0+ → bash hooks/local/upgrade.sh.
Routine CLI refresh → fusebase update --skip-skills.
Full CLI refresh → fusebase update, then bash hooks/local/post-fusebase-update.sh to restore the Flow overlay and wiring. Use only in an authorized/disposable target; never infer permission from this documentation.
.fusebase-flow-source/ is transient and can affect flat-config lint; see docs/fusebase-cli-edition.md for the supported cleanup/ignore path.

Project-specific values

For maintenance of fusebase-dev/fusebase-flow itself, read docs/maintainer-execution.md and follow its project-specific implementation, review and testing process. This exception does not apply to consumer app work or relax executable safety controls.

Field	Value	Where enforced
Project name	(run `/onboard` or edit)	informational
Stack	(run `/onboard` or edit)	informational
Workflow mode	`direct_to_main`	`policies/approval-policy.yml`
Worker-undisturbed paths	`none` (extend if needed)	`policies/protected-paths.yml`
Decision letter prefix	`B`	`templates/decisions.md`
T-counter	`11`	`templates/tasks.md`

Where Fusebase Flow and project-specific rules conflict, project-specific rules win.

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.

Check your setup, free