Skip to content

all-development-protocols.md

Latest commit

 

History

History
79 lines (68 loc) · 5.83 KB

all-development-protocols.md

File metadata and controls

79 lines (68 loc) · 5.83 KB
Raw
name protocol:all-development-protocols
description Router for shared development protocols: read order and file roles for the protocol folder.
date 09-06-26
metadata
node_type type read_order required read_when
memory
protocol
0
true
any substantial planning, research, validation, or implementation task — read first to route into the protocol folder

Development Protocols

Canonical shared workflow rules for this repository live here.

Use this folder for durable, repo-specific operating instructions that must stay aligned across Claude, Codex, and future agent systems. Keep tool-specific bootstrapping in AGENTS.md and CLAUDE.md, but keep the actual shared protocol content here.

Read Order

  1. vc-system-behavior/ — complete system behavior reference, 12-file split (start here: vc-system-behavior/01-overview.md is the entrypoint). The legacy 186KB monolith is archived at archive/vc-system-behavior-reference_ARCHIVED_09-06-26.md (read-only history).
  2. orchestration.md
  3. implementation-standards.md
  4. plan-lifecycle.md
  5. phase-programs.md
  6. context-maintenance.md
  7. communication-standards.md
  8. vc-autoresearch-spec.md (optional deep reference — read only when designing or deeply understanding the autoresearch gap-loop primitive)
  9. autopilot.md — Autopilot Mode trigger phrases, consolidated clarification, provisional goal block format, mode markers, per-gate decision policy, hard stops, and deactivation rules. Read when: user says autopilot / autonomous mode / /autopilot trigger phrases, or when understanding orchestrator §Autonomy Mode integration.

File Roles

  • vc-system-behavior/ (12-file split — canonical live reference; 01-overview.md entrypoint) Authoritative end-to-end system behavior reference: every phase, every required skill (with REQUIRED vs CONDITIONAL labels), both test loops, /goal block format, phase program inner loop, infrastructure gaps (D1/H4/G1/I2), and before/after hardening table. Split into numbered files 01-overview … 12-reference. The legacy single-file monolith is archived (read-only) at archive/vc-system-behavior-reference_ARCHIVED_09-06-26.md.

  • orchestration.md Delegation rules, subagent status protocol, context isolation, feature-scope routing, intent clarification rules, validate gate and skip conditions, BLOCKED escalation path, two-tier fan-out escalation, and research-first rules for service-shaped work.

  • implementation-standards.md Durable implementation standards, file-size guidance, error-handling preferences, quality gates, and commit hygiene.

  • plan-lifecycle.md How plans are named, where they live, when to use feature folders, how EXECUTE handoff works, and how mixed legacy plan shapes should be treated.

  • phase-programs.md How to run large multi-phase programs: umbrella plan, per-phase plan split, required 10-step loop, blocker handling, and foundation-vs-expansion boundaries.

  • context-maintenance.md How process/context/ is organized, when to create or split groups, how to keep all-context.md accurate, and how long-lived knowledge differs from feature plans.

  • communication-standards.md How agents write human-facing output: answer-first (BLUF), plain language, TL;DR, no filler. Single source of truth for output style; CLAUDE.md, AGENTS.md, prose-producing agents, and report/spec/closeout/clarify skills point here instead of restating it.

  • vc-autoresearch-spec.md (optional deep reference; read_order: 7, required: false) Deeper design reference for the vc-autoresearch loop primitive (the find-gaps → fix → repeat bookkeeper wired into PVL and EVL): canonical domain configs, iteration-report frontmatter schema, and TSV log format. The operative contract is .claude/skills/vc-autoresearch/SKILL.md; this file is the design rationale behind it.

  • autopilot.md Autopilot Mode protocol: trigger phrase detection, consolidated clarification round (one-round rule), provisional goal block format (9 fields, ≤4000 chars, EXECUTE CONSENT required), [MODE: AUTOPILOT | X] marker syntax, [AUTOPILOT CONTEXT] injection schema for subagents, AUTOPILOT_ACTIVATED signal, V6/V7 (UPDATE) variant, per-gate autonomous decision policy, hard stops, deactivation rules, and phase-program interaction. Complements orchestration.md §Autonomy Mode.

  • references/program-goal-charter-template.md Blank and filled Program Goal Charter template for phase-program umbrella plans. Read when building an umbrella plan or needing the compressed session-goal block format. Canonical content also mirrored at .claude/skills/vc-generate-phase-program/references/program-goal-charter-template.md.

Note: Protocol procedures for intent clarification, parallel strategy scoring, validate fan-out, and closeout are now in the corresponding skills. See process/context/all-context.md skill registry for skill names (vc-intent-clarify, vc-agent-strategy-compare, vc-validate-findings, vc-generate-closeout, vc-risk-evidence-pack, vc-generate-phase-program).

Maintenance Rules

  • Update these files first when shared workflow behavior changes.
  • When AGENTS.md, CLAUDE.md, agent prompts, hook reminders, or skill guides describe shared repo workflow, they should point here instead of duplicating large blocks.
  • After changing protocol files, re-run the relevant validators and any hook tests that resolve rule paths.
  • Discovery frontmatter: every file in this folder (recursive, incl. vc-system-behavior/) carries leading YAML frontmatter (name: protocol:<slug>, description, date, metadata: {node_type, type: protocol, read_order, required, read_when}) so vc-context-discovery can route on it. note.md is the single intentional exception (raw scratch dump). Enforced by node .claude/skills/vc-audit-context/scripts/validate-protocol-discovery.mjs (run via vc-audit-context).