Skip to content

source-map.md

Latest commit

 

History

History
76 lines (54 loc) · 5.31 KB

source-map.md

File metadata and controls

76 lines (54 loc) · 5.31 KB
Raw

Source map — Fusebase Flow

This document records source boundaries for the Fusebase CLI provider layer. Canonical Fusebase Flow files remain clean-room original. CLI provider assets are copied as Fusebase Apps CLI provider assets and are intentionally kept outside canonical Flow roots.

Standard attestation

Designed after reviewing public AI coding workflow patterns. No third-party code, prompts, skill files, or hook scripts are copied.

This wording is used in:

  • All 34 canonical SKILL.md Clean-room note sections (flow-skills/<slug>/SKILL.md).
  • All 68 mirror SKILL.md files (regenerated from canonical via mirror-skills.sh).
  • templates/skill-template.md (substrate for future skills).
  • hooks/README.md (hook framework attestation).

Edition-specific copied assets

Asset family Source Destination Boundary
CLI provider skills Fusebase Apps CLI project template .claude/skills/<cli-skill>/, .agents/skills/<cli-skill>/ Provider/domain assets; not canonical Flow skills
CLI app agents Fusebase Apps CLI project template .claude/agents/app-*.md, .codex/agents/app-*.md Provider/domain agents; not canonical Flow role agents
CLI quality hooks Fusebase Apps CLI project template .claude/hooks/* Claude Code provider hooks; separate from Flow hooks/handlers/*

These copied assets are documented in docs/fusebase-cli-edition.md. Do not cite the standard Flow clean-room attestation as applying to the copied CLI provider assets.

Pattern categories considered (generic descriptions only)

The following design patterns are common to public AI coding workflow discussion. Fusebase Flow implements them as original content; specific vendor names are not advertised in the public template.

Pattern category How Fusebase Flow implements it
Always-on repository instructions AGENTS.md (portable baseline) + provider-specific compatibility files
Skill catalogs flow-skills/<slug>/SKILL.md with frontmatter + predictable section structure
Rules / Skills / Workflows / Hooks separation This repo's directory layout under ``
Plan-before-edit discipline FR-02 + implementation-planning skill + architect-escalation workflow
Repo-map / context-map onboarding repo-onboarding-context-map skill
Lint / typecheck loop FR-13 + pre-commit git hook
Lifecycle hooks 8 Python handlers under hooks/handlers/ reading a unified event schema
Pre-commit secret scanning hooks/git/pre-commit + policies/secret-patterns.yml
Project-trust gating Documented in .codex/config.toml.example and noted in hooks/README.md
Direct-to-main vs branch + PR mode approval-policy.yml: workflow_mode

What is NOT copied

  • No SKILL.md prose from any third-party project is copied into canonical Flow flow-skills/.
  • No hook handler code or shell scripts from any third-party project are copied into canonical Flow hooks/handlers/, hooks/shared/, hooks/git/, or hooks/local/.
  • No vendor configuration examples reproduced verbatim; all examples (.claude/settings.json.example, .codex/config.toml.example, etc.) are written from the public protocol shape, not copied from any vendor sample repo.
  • No prompt text, system prompt, or skill description from any third-party project.

Verification

The clean-room property is validated by:

  1. Original wording check — the public-template tree is verified against a word-boundary search for non-target tool names; expected result is zero matches.
  2. Standard wording presence - preflight.sh is configured to inspect skill frontmatter; manual review of clean-room notes confirms the standard wording in all 34 canonical + 68 Flow mirror SKILL.md files.
  3. Edition boundary check - CLI provider assets remain under provider surfaces and are not added to flow-skills/ or canonical Flow mirror manifests.
  4. License attestation - see docs/clean-room.md for the explicit clean-room statement for canonical Flow files.

Internal research notes (not published)

Any internal research notes that reference specific external projects belong outside this GitHub template. They live in operator-private notebooks or internal company docs, not in the public template tree. The published template uses generic wording only.

FR-26 context-compression discipline (clean-room)

The FR-26 context-compression discipline update — flow-skills/token-economy/SKILL.md § Context compression discipline, the /token-waste-audit command, and the large-output audit class in hooks/local/token-waste-audit.py — is clean-room original FuseBase Flow content. It implements FuseBase-native behavioral rules and deterministic, stdlib-only audit heuristics. No third-party code, prompts, skill files, command text, hook scripts, docs, benchmark language, or dependency is copied or vendored.

Last amended

2026-05-27 - Fusebase CLI edition source map; distinguishes canonical Flow clean-room files
              from copied CLI provider/domain assets.
2026-06-15 - FR-26 context-compression discipline (large-context/large-output) noted as
              clean-room original; no third-party dependency or content introduced.
2026-07-11 - `product-owner` skill bridge added as clean-room Flow content; count now
              34 canonical skills / 68 Flow mirror SKILL.md files.