[hmziqagent]

Closes out 100% of docs/runner-remaining-work.md (areas 1–5). Area 6 stays deferred by decision, as the tracker requires.

What landed

A1 — Parity closeout (52ffcc4)

• parity_failure asserts an exact non-zero exit code, not just cross-runner equality
• TokioRunnerOptions/builder mirrors the Duct options: configurable default timeout (no more hardcoded 60s) and toggleable command logging
• Parity rows for ConfiguredDuctRunner, the streaming path, and FakeRunner, with 4 intentionally-excluded rows enumerated in parity_tests.rs

A2 — Command-intent extensions, all four rulings recorded (2367b2d)

• stdin_bytes: Option<Vec<u8>> — implemented additively; included in FakeRunner matching
• timeout_policy — implemented; takes precedence over timeout, old payloads serde-default to runner default; excluded from FakeRunner matching
• Shell opt-in — rejected permanently (direct argv only; doc paragraph in policy.rs)
• Output disposition — rejected: grep evidence shows zero callers needing more than OutputMode
• Both new fields follow the full pattern: serde defaults in both hand-written halves, old-payload JSON tests, explicit FakeRunner rulings

A3 — Duct stdin error mapping (3e1dd3c, cb39a42)

• DuctRunner writes an owned stdin pipe explicitly; write failures map to Error::StdinFailed with kill-and-reap before returning, matching Tokio
• The tokio streaming path routes through the shared write_stdin so the contract holds unqualified on every path of both runners
• Parity test asserts the same classification across both runners

A4 — Diagnostics hardening (a6e3cfd)

• redact_logs default-true on both runners — logs never print unredacted args by default
• Command span wrapping success and failure paths on all three real-runner paths
• Sanitized env summaries (display_env) on completion events

A5 — Crate docs + tracker closeout (2e9c6c8)

• 14-row capability matrix across DuctRunner / TokioRunner / streaming / FakeRunner
• Ten compiling doctests incl. the Unix clear_env/PATH footgun, redaction, output limits
• Lossy-UTF-8 behavior and the direct-child-only cleanup policy stated in crate docs
• docs/runner-remaining-work.md updated: every item's final status with rulings and shas

Verification

• cargo test -p toride-runner --features 'duct-runner tokio-runner serde stream fake': 286 unit + 23 doctests pass (baseline 227 + 9)
• cargo fmt --check clean; clippy at the recorded 2-warning pre-existing baseline, zero new
• cargo doc --no-deps clean; cargo check --workspace --all-targets exit 0
• Each area audited by a fresh read-only reviewer against the tracker acceptance criteria; 1 finding found and fixed (cb39a42)