feat(config): config-driven tracker backend via rig.yaml task: block

[alex-mextner]

What

Wires task-cli's tracker backend selection to be config-driven from the repo's rig.yaml (CTO decision #4136.4: "for hyperide a LOCAL rig.yaml setting -> Linear, default -> GitHub Issues").

task-cli now reads its per-repo tracker backend from a flat task: block in the repo's committed rig.yaml — the single source of truth for the agent toolchain:

# rig.yaml (repo root)
task:
  backend: linear      # or github-issues (the default for every other repo)
  team: HYP            # -> linear.team
  # project: ""        # -> linear.project
  # repo: owner/name   # -> github.repo

Design

• DEFAULT = GitHub Issues. No task: block (or no rig.yaml) -> clean fall-through to github-issues.
• Cascade (later wins): defaults -> global -> rig.yaml task: -> task.yaml -> --config. A repo keeping a native task.yaml is untouched (it still wins over rig.yaml).
• Flat block translated to task-cli's own shape: team/project/repo are shorthands; full github:/linear:/enforce:/classify:/session:/projects: sections may be nested verbatim and deep-merge with the shorthands (order-independent; a true leaf collision is a hard error).
• Forward-compatible with rig-cli (rig.yaml is a foreign file rig owns/evolves):
  • a parse error or non-mapping root falls through to the github default (logged, never fatal);
  • an unknown sub-key under task: is warned-and-ignored, not a crash;
  • version is not passed through (rig.yaml has its own top-level version);
  • validate() type-guards a non-string backend into a clean ConfigError instead of an unhashable-type TypeError on foreign input.

Tests

tests/test_config.py: default=github, rig.yaml override=linear, missing/inert/broken/non-mapping rig.yaml=github, flat+nested merge for both linear and github (both orders), leaf-conflict rejected, unknown-key and stray version tolerated, malformed backend -> clean error. tests/smoke.sh covers the end-to-end rig.yaml -> linear/HYP selection. Full suite: 223 passing.

Reviewed across models before commit (review-before-commit gate); GLM seat was throttled (429) throughout, so the diff was reviewed on the Opus correctness seat across three rounds — each round's findings fixed (data-loss on flat+nested mix, forward-compat for unknown keys, foreign-file tolerance, non-string backend guard).

🤖 Generated with Claude Code

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 46143068af

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".