docs+fix: ATLAS-CONTRACT.md v1.1.0 + fix two wrong atlas calls in doctor.zsh

[Data-Wise]

Summary

• Task 1: Bump docs/ATLAS-CONTRACT.md from v1.0.0 to v1.1.0, documenting the 5 new atlas v0.9.3 integration flags and correcting the output-format section from a false universal-flag claim to a per-command support matrix.
• Task 2: Fix two out-of-spec atlas calls in commands/doctor.zsh that would silently fail for any user with atlas installed (OOS-1: atlas config get backend → atlas config show; OOS-2: atlas mcp status → command -v atlas-mcp).

Changes

docs/ATLAS-CONTRACT.md (v1.0.0 → v1.1.0)

• Version Compatibility table: added row for flow-cli v7.10.x ↔ atlas v0.9.3 with a summary of the 5 new flags.
• Parties line: updated to v7.10.x / v0.9.3.
• Warm Path Commands table: added atlas session status row (previously absent entirely); added atlas trail --limit <n> row.
• Required Commands table: added atlas inbox --count row alongside existing atlas inbox.
• Output Format Specifications: replaced the incorrect "4 formats via a universal --format flag" claim with a per-command support matrix:
  • project list: table | json | names
  • project show: table | json | names | shell
  • session status: table | json (NEW)
  • stats: table | json | text | md
  • session export: ical | json
  • plan: --json flag (not --format)
• New contract sections: detailed output semantics for each of the 5 new flags (session status --format=json, project list --count, project list --suggest, inbox --count, trail --limit).

commands/doctor.zsh

• OOS-1 (atlas config get backend): atlas config has no get subcommand. The correct command is atlas config show, which emits JSON containing a storage field (the backend). Fixed to atlas config show | jq -r '.storage' with a grep/sed fallback when jq is absent. Diagnostic remains non-fatal.
• OOS-2 (atlas mcp status): atlas has no mcp subcommand. The MCP server is a separate binary atlas-mcp (declared in atlas's package.json bin). Fixed to command -v atlas-mcp. Message updated to "atlas-mcp binary available" vs "(optional MCP server, not installed)".

Decision recorded

The task brief noted these as pre-approved fixes. No behaviour changes to flow-cli's hot-path; both doctor checks are non-fatal diagnostics.

Test plan

• source flow.plugin.zsh && doctor — atlas integration section should show backend from config show instead of erroring
• doctor with atlas installed but jq absent — sed fallback produces same result
• doctor with atlas not installed — section still shows "optional, not installed" gracefully
• doctor output references "atlas-mcp binary available" / "optional MCP server, not installed" (no longer calls atlas mcp status)
• ./tests/run-all.sh — no regressions (doctor tests are non-interactive, should still pass)

https://claude.ai/code/session_01FuHPHhpeCDgcowKPnG2F4V

---

Generated by Claude Code

[Data-Wise]

Fixes pushed (e7eb6cb7) + adversarial review notes

Ran an adversarial review of this branch and pushed fixes directly to it. First verified the two doctor.zsh OOS fixes against atlas source — both are correct, so I left that logic untouched:

• OOS-1 ✅ — atlas config show does emit JSON (bin/atlas.js:1122 → console.log(JSON.stringify(cfg, null, 2))) and .storage is the backend (filesystem | sqlite). atlas config get genuinely doesn't exist, so the old call was broken as you described.
• OOS-2 ✅ — atlas-mcp is a separate binary (package.json bin), so command -v atlas-mcp is the right check (it's an on-demand stdio server — "installed" is the sensible signal).

What I changed

1. False-positive "connected" (pre-existing) — doctor printed ✓ atlas connected whenever command -v atlas succeeded, so a broken/half-installed atlas (e.g. a bad interpreter shebang) reported as connected. Now gated on atlas config show actually returning output. Reproduced + fixed against a real broken local atlas:

   ✓ atlas installed (vunknown)
   ⚠ atlas installed but not responding (config show failed)   ← was "✓ connected (filesystem backend)"
   ✓ atlas-mcp binary available
   

2. Hardened the no-jq .storage parse — anchored on the "storage" key. The prior sed 's/.*: *"..."/\1/' greedily grabs the last quoted value, so on compact JSON it returns the wrong field ({"storage":"filesystem","other":"db"} → db). Doesn't bite today (atlas pretty-prints) but it's a latent trap.
3. Documented atlas config show in ATLAS-CONTRACT.md — the code depends on it, but the contract only covered session/list/inbox/trail. Added the JSON shape, .storage, and the "no config get" note.
4. tests/test-doctor-atlas-calls.zsh — static spec-compliance guard (needs no working atlas or jq; runs on any CI runner), registered in run-all.sh. 5/5.

Heads-up

• This branch is based on old main (477b5c59), so it predates the CI-gate test helpers (skip_without_warm_atlas, run-all.sh rc-77) now on dev — that's why the new test is fully self-contained. Worth a rebase onto dev before merge.
• It's your PR — left it as a draft for you to mark ready / merge.

🤖 Generated with Claude Code