Skip to content

ROADMAP.md

Latest commit

 

History

History
85 lines (61 loc) · 4.14 KB

ROADMAP.md

File metadata and controls

85 lines (61 loc) · 4.14 KB
Raw

atlas roadmap

Organized by user outcome, not by component — what each pile of work is for. Issues are linked under the outcome they serve; this is a guide, not an exhaustive list of the issue tracker.

atlas is alpha. The line below marks what's shipped vs. planned; the alpha → beta → stable bar is defined in docs/release-readiness.md, and what's a stable contract vs. what may still change is in docs/stability.md.

Legend: ✅ shipped · 🚧 in progress · ⬜ planned.

Trust the output

The benchmark claim must be believable and the formats must be dependable.

  • ✅ Reproducible comprehension benchmark (~65% fewer tokens at equal accuracy) — benchmark/README.md
  • ✅ Machine-readable JSON Schemas for map and diff — #57, #58
  • ✅ Pre-1.0 stability & deprecation policy — #97
  • ⬜ Public benchmark case study for the README claim — #40
  • ⬜ Per-language extraction-accuracy report — #111, backed by a real-world fixture corpus — #110
  • ⬜ Competitive benchmark arms (Aider repo-map / ctags / file-tree at equal budget) — #13
  • ⬜ Large-repo performance matrix — #73

Install anywhere

Make atlas trivial to get, on any platform and in locked-down environments.

  • ✅ pip/pipx (atlas-map), prebuilt binaries (macOS/Linux/Windows), cargo install from source
  • ✅ Offline / proxy / air-gapped install guidance — docs/install-offline.md
  • 🚧 PyPI Trusted Publishing — #29
  • ⬜ crates.io publish — #37 · Homebrew — #38 · Docker image — #67 · npm/npx wrapper — #51
  • ⬜ Packaged shell completions — #61 · man page — #62
  • ⬜ Supply chain: checksums & signatures — #63 · SBOM — #66 · SLSA provenance — #65
  • ⬜ Install smoke tests — #43 · binary-size/startup tracking in CI — #114

Drop into any agent

The map is only useful if it reaches the agent with minimal friction.

  • ✅ Agent integration cookbook — #39 · CLAUDE.md/AGENTS.md snippets — #87 · editor task snippets — #88
  • ✅ atlas-orient agent skill — skills/
  • ✅ MCP stdio server (get_map, get_symbol, …) — docs/CLAUDE_CODE_MCP.md
  • ✅ Budget presets (--preset) — #104
  • 🚧 atlas doctor install/repo diagnostics — #47
  • ⬜ MCP conformance tests against real clients — #101
  • ⬜ Selectable tokenizer profiles (Claude/OpenAI/generic) — #103
  • ⬜ Reusable GitHub Action — #68 · atlas init scaffolding — #52 · config file (.atlas.toml) — #59

Review & edit workflows

Help agents (and humans) with change-centric tasks, not just first-look orientation.

  • ✅ Structural atlas diff (trees or git revisions) + --exit-code CI gate
  • ✅ atlas --check committed-map freshness gate
  • ⬜ Git-aware focus from changed files — #105 · changed-map mode (files + dependents) — #108
  • ⬜ PR summary from diff — #70 · PR comment mode — #86
  • ⬜ Warning mode for unresolved import edges — #93 · compression-ratio report — #74

Cover your languages

  • ✅ Python, TypeScript/JavaScript, Rust, Go, Java, C/C++ — see the language matrix
  • ⬜ Tier 2 signature-accuracy audit — #22
  • ⬜ Framework-flavored examples (no semantic analysis) — #112
  • ⬜ Language-request template + prioritization rubric — #109

Live context

  • ⬜ --watch daemon for incremental re-mapping — #8, with a concrete behavior contract — #28

Core map quality (benchmark-gated)

Ranking/budgeting changes ship only with a measured win — see docs/SELF_IMPROVEMENT.md.

  • ✅ Collapsed-tail symbol index (ADR 0004) · stable anchors / progressive disclosure (ADR 0009)
  • ⬜ Better default-budget ranking for known comprehension misses — #30
  • ⬜ Evaluate adaptive default budget — #79
  • ⬜ Grow the benchmark suite to 10 balanced tasks — #6

This roadmap is updated as major issues close. Outcomes are stable; the issues under them will churn. For day-to-day status see STATUS.md; for shipped changes see CHANGELOG.md.