Add direct workflow dry-run contract

[wren-nutthouse]

Summary

Milestone A only from the Tutti fix-up plan: establish the direct workflow dry-run/config contract without implementing any provider/API execution.

• Adds tt run --direct as a planning/dry-run mode selector for prompt steps.
• Extends prompt workflow config with optional direct, provider, model, and policy metadata.
• Resolves direct prompt steps to an explicit internal direct step shape.
• Shows direct provider/model/policy/prompt metadata in human and JSON dry-run output.
• Rejects non-dry-run direct execution with an actionable “not implemented yet” error.
• Preserves existing non-direct workflow paths by defaulting all service/webhook/verify/scheduler call sites to direct: false.

Non-goals

• No real model/provider/API execution.
• No provider secrets or global-provider lookup.
• No dashboard/product redesign.
• No atomic-write/env-scrub/replay/context-compaction work.
• No merge or modification of existing open PR branches.

Verification

Local verification from /Users/adamnutt/work/nutthouse/tutti:

cargo fmt
cargo test --quiet
# 463 unit tests passed
# 1 integration test passed

cargo clippy --all-targets -- -D warnings
# finished successfully

cargo build --quiet
# finished successfully

Dry-run smoke checks:

(cd /tmp/tutti-direct-smoke && target/debug/tt run plan --direct --dry-run --json)
# emitted a `type: direct` step with provider/model `default/default`, policy `read_only`, and output_json path

(cd /tmp/tutti-direct-config-smoke && target/debug/tt run plan --dry-run)
# emitted human dry-run row: type direct, provider openai, model gpt-5.6-sol, policy read_only

(cd /tmp/tutti-direct-config-smoke && target/debug/tt run plan)
# exits non-zero with: direct execution is not implemented yet; use `tt run plan --dry-run`

Open PR signal reviewed before opening this PR

Current open PRs and interaction risk:

PR	Signal	Interaction with this PR
#130 Finish Tutti trust and activation sprint	CONFLICTING / DIRTY, review required. Touches activation/runs/automation/docs.	Overlaps src/automation/mod.rs, src/cli/mod.rs, src/main.rs; should not be merged without rebasing after this or vice versa.
#132 Add configurable tmux/Zellij multiplexer backends	Mergeable but blocked on review/admin; CodeRabbit latest check green and author reports local cargo test --quiet pass.	Overlaps src/automation/mod.rs, src/config/mod.rs, src/main.rs; likely needs a rebase if this lands first.
#133 Add OpenCode CLI runtime adapter	Mergeable but review required; Greptile flagged broad "opencode" idle pattern and missing idle/completion/auth tests.	Overlaps src/config/mod.rs; direct contract should compose, but #133 should address Greptile adapter-detection findings before merge.

Product/market-learning signal folded in

Chief-of-Staff / operating-context learnings point Tutti toward an evidence/control-plane product, not just another runtime adapter collection:

• Direct API execution needs explicit provider/model/policy and run-id correlation before real execution.
• The market direction is orchestration with auditable gates, recovery, evidence, and multi-agent SDLC loops: Claude/CE for planning/review/ops, Codex/CE for bounded implementation/testing, and proof before merge.
• Runtime breadth matters (Add configurable tmux/Zellij multiplexer backends #132/feat(runtime): add OpenCode CLI runtime adapter #133), but it should not outrun the control-plane contract: dry-run inspectability, explicit policies, no secret leakage, and non-ambiguous execution ledgers.

This PR intentionally ships only the contract/dry-run seam so real direct execution can be reviewed as a separate provider/run-ledger design.

[greptile-apps]

Greptile Summary

This PR adds a direct workflow planning contract without enabling provider execution. The main changes are:

• Adds tt run --direct for direct prompt dry-runs.
• Adds provider, model, policy, and direct metadata to prompt configuration.
• Preserves prompt inputs and controls when resolving direct steps.
• Shows direct step details in human and JSON dry-run output.
• Rejects direct execution outside dry-run mode.

Confidence Score: 5/5

This looks safe to merge.

• Direct steps retain configured prompt inputs and controls.
• Unsupported policy values are rejected during configuration validation.
• Direct dry-runs now apply the artifact prerequisite check.
• No blocking issues remain in the reviewed changes.

Important Files Changed

Filename	Overview
src/automation/mod.rs	Adds direct step resolution, preserves prompt inputs and controls, and blocks direct execution.
src/config/mod.rs	Adds direct prompt metadata and validates required provider, model, and policy values.
src/cli/run.rs	Adds direct dry-run rendering, JSON serialization, and artifact prerequisite validation.
src/cli/mod.rs	Adds the --direct argument to the run command.
src/main.rs	Forwards run options through the new structured request.

_{Reviews (3): Last reviewed commit: "fix: validate direct artifact dry-run pr..." | Re-trigger Greptile}