docs: retire Maestro compatibility tracker

[thymikee]

Summary

Retire the stale #558 Maestro compatibility umbrella while keeping a concise versioned capability and limitations reference in agent-device help maestro and the website docs.
Treat conformance divergences as a mechanical parity record, with focused issue links only when implementation work is planned, and preserve #1342's intentional we-are-lenient cases.

Validation

• pnpm format:check
• pnpm typecheck
• pnpm lint
• pnpm exec vitest run src/compat/maestro/__tests__/support-matrix.test.ts src/cli/parser/__tests__/cli-help-command-usage.test.ts src/cli/parser/__tests__/cli-help-topics.test.ts (61 tests)
• pnpm maestro:conformance (46 tests)
• Broad related run: 863/868 passed in parallel; all five fixed-timeout failures passed when their four files were rerun with one worker (23 tests)

[thymikee]

Two blockers before this can be ready:

1. The branch conflicts with merged feat(maestro): numeric option fields accept ${VAR} lookup interpolation (#1293) #1342 in scripts/maestro-conformance/expected-divergence.ts. Rebase while retaining feat(maestro): numeric option fields accept ${VAR} lookup interpolation (#1293) #1342’s we-are-lenient model and removing the obsolete upstream/042_extended_wait divergence; otherwise the tracker cleanup would restore a divergence feat(maestro): numeric option fields accept ${VAR} lookup interpolation (#1293) #1342 made identical. As part of that reconciliation, remove the completed Maestro compat: numeric option fields accept ${VAR} lookup interpolation (extendedWaitUntil.timeout, tapOn.index, siblings) #1293 links from the four new leniency records if tracking should exist only for planned work.
2. The PR deletes the only versioned user-facing supported-subset and operational-boundary guidance, replacing it with “focused subset” plus ADR 0015. That ADR does not enumerate supported commands/grammar/platform constraints, env precedence, or the runScript trust/network caveat. This conflicts with the repository rule that versioned CLI help is the agent-facing source of truth, and the weakened tests no longer protect those claims. Preserve a concise versioned capability/limitations reference (or an equivalent generated user-facing source) and test substantive coverage.

The current head has no full post-rebase CI evidence.

[github-actions]

PR Preview Action v1.8.1
Preview removed because the pull request was closed.
2026-07-20 19:09 UTC

[github-actions]

Size Report

Metric	Base	Current	Diff
JS raw	1.8 MB	1.8 MB	+1.1 kB
JS gzip	574.2 kB	574.5 kB	+382 B
npm tarball	688.6 kB	689.0 kB	+424 B
npm unpacked	2.4 MB	2.4 MB	+1.1 kB

Startup median (7 runs, lower is better):

Scenario	Base	Current	Diff
CLI --version	29.4 ms	28.1 ms	-1.3 ms
CLI --help	62.6 ms	62.1 ms	-0.5 ms

Top changed chunks:

Chunk	Raw diff	Gzip diff
dist/src/cli-help.js	+2.1 kB	+906 B
dist/src/registry.js	-1.0 kB	-524 B

[thymikee]

Re-reviewed 6e30969d: both prior blockers are resolved. The branch now cleanly preserves #1342’s we-are-lenient model, keeps the obsolete extended-wait divergence removed, and drops completed #1293 tracking links. Substantive versioned guidance is restored through agent-device help maestro, shared capability/limitation data, website parity, and tests covering commands, platform/launch boundaries, environment precedence, and the runScript trust/network caveat. CI is fully green; no remaining findings.