Restore the approved README and require approval for edits

[arjitj2]

Closes #154

Restores README.md exactly to its state before #153, as requested by the owner. Adds an explicit approval requirement for future README edits in AGENTS.md. The six-hour adoption task now carries the same restriction.

The Setup correction, release 1.17.4 and installed package remain unchanged. The restored wording is the prior owner-approved copy, not a new assessment of upstream capabilities.

Validation: exact README byte equality against 26f4aff^:README.md; unchanged plugins/pstack tree; documentation, maintenance, hooks, ledger, static invariants and Claude plugin validation pass. Strict typechecking passes. The local Node 24 suite failed in unchanged tests with 990 passes, eight failures and seven additional errors; #141 tracks this test problem. Exact-head Linux CI passed with normal budgets. The actual scheduled task was updated through the app API and read back, with its schedule and active status preserved.

Related ongoing monitor #1. Fixture work remains separate in #141.

Copilot review was requested through the CLI and API, but no request persisted. It is unavailable, not an approval. Independent exact-head shipping review passed with notes.

[arjitj2]

Independent Shipping verdict: PASS+NOTES

PR: #155

• Reviewed head: 6667c727f915d8898c4756f6744c30e08af852cc.
• Patch base: 26f4aff278307fe8b016eb13a911978586171906.
• Stable patch ID: 5b0cd4b060b2c3ce1d31cf369cf5c8e2a4ca3942.
• I did not implement these changes. My only write was this report.

Verified

The entire diff changes only root README.md and AGENTS.md. README at the reviewed head equals 26f4aff^:README.md byte for byte. Its SHA-256 is 98a571cf5ee367259b267f81720e0128b8237ed7c6ae1fda0a3e0a36d93530cf. This exactly restores the two README passages changed by PR153 without reverting its Setup mapping, provenance, or version metadata.

The new AGENTS rule requires explicit owner approval before any README edit, including positioning and purpose prose. It requires proposed diffs outside the repository and limits this rollback authorization to this rollback. This matches the user's instruction.

I directly read the active automation TOML. Its prompt has the same approval guard, explicitly excludes routine catch-up authority for README changes, and preserves independently authorized work. It retains ACTIVE status, the existing target thread, and FREQ=DAILY;BYHOUR=1,7,13,19;BYMINUTE=45;BYSECOND=0. This readback agrees with verification.json. I did not perform or independently observe the earlier app update transaction.

The package tree is unchanged at dd102020515e452f65d78078cbeb24cdcc549e41 in both base and head. No release or installation update is required. Version remains 1.17.4. The worktree is clean and its HEAD matches the reviewed head.

I independently ran python3 scripts/check-docs.py; it passed. I inspected the retained local logs and exit codes: strict typing, documentation, 27 maintenance tests, 8 hook tests, 16 docs tests, ledger coverage for 48 entries, static checks, and plugin validation passed. The local Bun suite failed: 990 passes, 8 failures, and 7 additional errors. These are not represented as passing, and this review does not establish their cause. No runtime or test files changed in this patch.

Source/story check: UPSTREAM.md explicitly excludes upstream's generated Claude package and incompatible adapter rules. PR153 imported no generated packaging; its functional skill change was the portable Setup question-tool name. Its README rewrite was a separate descriptive change. Restoring historical README prose here is the owner's requested rollback, not a new verification or endorsement of its current upstream comparison.

Remaining landing gate and limits

At inspection, exact-head CI39046964510 was IN_PROGRESS for the reviewed head. PASS+NOTES is an independent patch verdict, conditional on the parent obtaining final successful required CI before landing. It is not a claim that CI is already green. The parent must refresh head/base, check status and review findings immediately before a guarded merge.

My GraphQL read confirmed both autoMergeRequest and mergeQueueEntry null, head unchanged, and base main. I did not arm or modify any queue, publish, merge, release, or alter installation state. No live model-host checks were repeated because the package tree is unchanged; no GUI behavior was exercised. No implementation issue is complete until the parent verifies merge and closure.