Generated Files
Contents
- Purpose
- Minimal-Footprint Rule
- Root Files
- GitHub Copilot Files
- Claude Files
- Codex Files
- .vibeflow Files
- Dispatch Files
- Hook Files
- Source Modules That Materialise the Generated Files
Purpose
This document describes files the orchestrator should generate in a target repository.
Minimal-footprint rule
The file lists below are the maximum surface VibeFlow may generate, not a mandatory
set. Per the minimal-footprint and AI-generated-output principle in MASTER_SPEC.md, the
orchestrator must:
- generate only the files the selected engine and current task actually require
- skip optional/empty artifacts instead of creating placeholder files
- compose every file from canonical context with the AI at runtime (no static templates)
- keep generated files disposable and reproducible from the canonical .vibeflow/* source
Root files
CLAUDE.md
Primary instruction file for Claude Code.
Should contain:
- project overview
- architecture
- build/test/lint commands
- coding conventions
- workflow policy
- skill usage policy
- security rules
- verification rules
- skill evolution policy
AGENTS.md
Cross-tool instruction file for Codex, Copilot, and other coding agents.
Should contain:
- shared project context
- agent orchestration policy
- skill usage policy
- anti-hallucination policy
- verification requirements
- task execution rules
GitHub Copilot files
.github/copilot-instructions.md
.github/instructions/backend.instructions.md
.github/instructions/frontend.instructions.md
.github/instructions/testing.instructions.md
.github/instructions/security.instructions.md
Purpose:
- repo-wide Copilot instructions
- path-specific coding rules
- testing/security guidance
Claude files
.claude/agents/
.claude/skills/
.claude/settings.json
Example agents:
orchestrator.md
backend-engineer.md
frontend-engineer.md
test-engineer.md
security-reviewer.md
devops-engineer.md
skill-compliance-reviewer.md
Example skills:
repo-onboarding/SKILL.md
skill-maintainer/SKILL.md
markdown-reader/SKILL.md
docx-reader/SKILL.md
xlsx-reader/SKILL.md
pptx-reader/SKILL.md
pdf-reader/SKILL.md
context7-docs/SKILL.md
Codex files
.codex/config.toml
.vibeflow/dispatch/codex.md
Purpose:
- standard Codex execution profile
- prompt template generated from canonical context
.vibeflow files
.vibeflow/PROJECT_CONTEXT.md
.vibeflow/REQUIREMENTS.md
.vibeflow/TASK_CONTEXT.md
.vibeflow/WORKFLOW_POLICY.md
.vibeflow/SKILL_INDEX.md
.vibeflow/SETTINGS.json
.vibeflow/WORKFLOW_STATE.json
.vibeflow/workunits/<name>/CONTEXT.md
.vibeflow/workunits/<name>/evidence/
.vibeflow/attachments/ # uploaded sample files (web UI)
.vibeflow/ai-context/ANTHROPIC_SKILL_STANDARD.md # copied from src/skills/ on init
.vibeflow/ai-context/SKILL_TAXONOMY.md # copied from src/skills/ on init
.vibeflow/skills/<name>/ # canonical skill store (mirrored to engines)
SKILL.md # required
references/ # optional
scripts/ # optional
assets/ # optional
Purpose:
- canonical workflow memory
- normalized context
- skill registry summary
- per-repo settings: optional tool toggles (codegraph/lsp) and tool priority
- current task state
- per-task work units, gates, and resource ledger (see WORK_UNIT_ORCHESTRATION.md)
SETTINGS.json is written on every vf init from DEFAULT_SETTINGS. Each work unit dir
holds CONTEXT.md (the dispatch prompt) and an evidence/ folder of JSON results
(<engine>.result.json, investigation.json); per-unit state lives centrally in
WORKFLOW_STATE.json, not in a per-unit meta.json. TODO.md and HANDOFF.md are planned
but not yet generated (see WORK_UNIT_ORCHESTRATION.md).
Work-unit files follow the minimal-footprint rule: create them only when a task is
decomposed (3+ files across multiple modules, or delegated agents). See
WORK_UNIT_ORCHESTRATION.md for when units are created and how they are tracked.
Dispatch files
.vibeflow/dispatch/claude.md
.vibeflow/dispatch/codex.md
.vibeflow/dispatch/copilot.md
Each dispatch file should include:
- task goal
- selected context
- selected skills
- allowed files
- constraints
- verification commands
- expected output schema
Hook files
vf hooks emit writes per-engine native hook configs into the target repo. There are no
per-event .vibeflow/hooks/*.ts scripts; instead every config delegates to the single
vf hook entrypoint (reads a JSON event on stdin, returns a decision):
.claude/settings.json (Claude PreToolUse/PostToolUse/Stop → `vf hook`)
.codex/hooks.json (Codex post-command/post-write/verify-result → `vf hook`)
.github/hooks/copilot.json (Copilot preToolUse (fail-closed) + postToolUse → `vf hook`)
.githooks/pre-commit (fail-closed shell hook routing staged files → `vf hook`)
Purpose:
- enforce permissions (Claude: native pre-action block; Codex/Copilot: detection-only)
- reduce unsafe actions
- verify skill usage
- validate output
- reduce false positives through risk scoring
Source modules that materialise the generated files
The runtime source-of-truth modules for the generated surface (under src/):
src/ai-init.ts # writes PROJECT_CONTEXT, REQUIREMENTS, SKILL_INDEX, dispatch prompts
src/skills/sync.ts # canonical → mirror sync (.vibeflow/skills → .claude/.agents/.github)
src/skills/validator.ts # Anthropic skill-creator standard validation
src/skills/importer.ts # Context7 + local-dir skill import (temp → validate → promote → sync)
src/skills/ANTHROPIC_SKILL_STANDARD.md # reference, copied to .vibeflow/ai-context/ on init
src/skills/SKILL_TAXONOMY.md # reference, copied to .vibeflow/ai-context/ on init
src/probe-cache.ts # 60s stable / 5s short-TTL engine readiness cache
src/engine-quota.ts # parse claude/codex/copilot quota output for the preflight gate
src/preflight-delegate.ts # 3-layer gate: presence → auth → quota, with auto-fallback
src/memory/ # MemoryProvider seam + builtin (bun:sqlite FTS5) + claude-mem providers
Runtime-generated files (gitignored)
.vibeflow/knowledge/memory.db— SQLite FTS5 index ofdecisions.md(builtin memory mode; regenerated on mtime change; gitignored)WORKFLOW_POLICY.md— whenvf init --memoryaccepted: the## Memory: claude-memguide block is appended
Related: Architecture · Tool Adapters Edit this page on GitHub