Skip to content

CLAUDE_INDEX.md

Latest commit

 

History

History
80 lines (64 loc) · 3.63 KB

CLAUDE_INDEX.md

File metadata and controls

80 lines (64 loc) · 3.63 KB
Raw

CLAUDE_INDEX.md — Thin Always-Loaded Index (§11.4.141)

Purpose: ~5-8K token entry point for fresh sessions. Full rules in constitution/Constitution.md. Session state: .remember/remember.md (§11.4.131)

Essential Rules (one line each, literal anchors for propagation gates)

  • No CI/CD pipelines — no .github/workflows/, .gitlab-ci.yml, Jenkinsfile (§11.4.156)
  • No mocks in production — mocks/stubs only in unit tests (§11.4.27/CONST-050)
  • SSH-only Git — git@github.com:… URLs, never HTTPS (CONST-043)
  • Real data for non-unit tests — real DB, real HTTP, real containers (§11.4.27)
  • 100% challenge coverage — every component needs Challenge scripts (CONST-048)
  • Reproduction-before-fix — reproduce bug BEFORE fixing (§11.4.43)
  • Definition of done — pasted terminal output from real run required (§11.4.5)
  • No self-certification — "verified/tested/working" forbidden without evidence (§11.4.6)
  • Zero-bluff mandate — tests must guarantee Quality+Completion+Usability (CONST-035/§11.4)
  • No force-push — merge-onto-latest-main, never force (§11.4.113)
  • No hardcoded content — all user-facing text from LLM/i18n/metadata (CONST-046)
  • Submodule-dependency manifest — every submodule ships helix-deps.yaml (CONST-054)
  • Fetch-before-edit — git fetch --all FIRST in every session (CONST-060)
  • Subagent-driven default — fresh implementer per task (§11.4.70)
  • Zero-idle — never sit idle if work is progressable (§11.4.94)
  • CodeGraph mandatory — codegraph init + index for all AI-agent projects (§11.4.78)

Build & Test Commands

# Root governance gates (from repo root)
make no-silent-skips    # fail on bare t.Skip()
make demo-all           # run every submodule's demo
./setup.sh              # first-time: submodules + deps + build

# Inner application (from helix_code/)
make build              # → bin/helixcode
make verify-compile     # quick compile sanity
make test               # all unit tests
make test-coverage      # coverage with -race
make dev                # build + run with dev config

# Full integration (real PG + Redis + Ollama)
make test-infra-up      # start docker-compose stack
make test-full          # ALL tests, ZERO skips
make test-infra-down    # tear down

# Anti-bluff smoke
grep -rniE "\bsimulated\b|\bfor now\b|TODO implement" helix_code/internal helix_code/cmd | grep -v _test.go | grep -q . && echo "BLUFF" || echo "clean"

Anti-Bluff Checklist (§5)

  • No simulation/placeholder in production code
  • Real HTTP calls (not mocked)
  • Real database operations
  • Real process execution (os/exec)
  • Tests validate reality, not function call counts
  • Challenge validates end-to-end
  • No bare t.Skip() without SKIP-OK marker
  • Evidence pasted in commit/PR

Technology Stack (§3.1)

Go 1.26 (inner app), Gin v1.11.0, PostgreSQL 15+, Redis 7+, golang-jwt/v4, Viper v1.21.0, Cobra v1.8.0, AWS Bedrock, Azure azcore, Fyne v2.7.0 (desktop), tview (TUI), stretchr/testify.

Repo Layout (§3.2)

helix_code/          # meta-repo root (governance + submodules)
├── helix_code/      # inner Go application (dev.helix.code)
├── submodules/      # owned submodules (per .gitmodules)
├── constitution/    # constitution submodule (source of truth)
├── dependencies/    # third-party submodules
├── docs/            # architecture, QA evidence, research
├── scripts/         # init, governance, gates
└── .remember/       # session resumption (§11.4.131)

Current State

See .remember/remember.md for live HEAD, branch, queue, and execution plan.