macOS: run execute doctor node fails 'bootstrap readiness failed' — installer does not deploy bin/legacy + bin/pipeline.config.json

[lunnynight]

Problem

On macOS, code-intel run execute fails at the doctor node with domain_failed, diagnostic doctor diagnosis: bootstrap readiness failed, exit code 10.

Root cause: the doctor node resolves the legacy runner script and pipeline config via paths.bin → ~/Library/Application Support/code-intel/bin/:

• bin/legacy/run-code-intel.ps1 — must exist
• bin/pipeline.config.json — must exist

The installer/bootstrap deploys the release binary, the sentrux shim, and orchestration/, but does not deploy these two items into bin/. Doctor's auto-resolved path is fixed at bin/ (not CODE_INTEL_HOME/release root), so passing --config to a different location does not help.

Impact

• Every run execute on macOS reports domain_failed even though all other nodes pass (9/9 nodes succeeded in our run; only doctor failed).
• Artifacts from domain_failed runs are NOT treated as query authority — artifact query rejects them (index diagnostic non_completed). So the pipeline is unusable for artifact query / change impact on macOS until this is fixed.

Repro

code-intel doctor bootstrap --repo <alias> --repo-path <repo> --json
# → ok:false, missing: ["pipeline script", "pipeline config"]

Workaround (verified 2026-08-08)

Symlink the release assets into bin/ (release root stays the single source of truth):

ln -s "$HOME/Library/Application Support/code-intel/releases/v0.7.0/legacy" "$HOME/Library/Application Support/code-intel/bin/legacy"
ln -s "$HOME/Library/Application Support/code-intel/releases/v0.7.0/pipeline.config.json" "$HOME/Library/Application Support/code-intel/bin/pipeline.config.json"

After the symlinks: doctor bootstrap returns ok:true, missing:[], and run execute completes (exit 0, 9 nodes pass, 1539 anchors verified).

Suggested fix

Either (a) have the bootstrap/installer deploy legacy/ + pipeline.config.json into bin/, or (b) make the doctor resolve these from the release root / CODE_INTEL_HOME instead of a fixed bin/ path on macOS.

Priority: P2 — blocks macOS users from producing query-authority artifacts.

[lunnynight]

Related finding: doctor graphProvider check resolves a source-build path that the release install never creates

Even after the bin/legacy + bin/pipeline.config.json wiring above is fixed, doctor bootstrap reports:

"graphProvider": {
  "binaryFound": false,
  "cargoFound": false,
  "sourceFound": false,
  "command": "<CODE_INTEL_HOME>/bin/target/release/code-intel graph --repo <repo-path> --language zh --write --json"
}

The check looks for a source build (bin/target/release/code-intel) under the bin dir. The released installer ships only the compiled binary (and symlinks bin/code-intel → release binary), never a target/ checkout — so binaryFound is always false on a release install even though code-intel graph --repo <path> --write works fine from the release binary.

Impact: graph evidence is permanently reported graph_missing / graph.code-intel: missing (doctor observation understandGraph: false, repowiseState: false) on every release install, regardless of whether the graph was actually produced.

Suggested fix: the doctor should resolve the graph producer to the installed binary (the same one that resolves bin/code-intel), not a target/release source build. Or treat binaryFound via the bin/code-intel entrypoint.