docs(agents): a green check is evidence only once you have seen it red - #1547
Conversation
Three vacuous regression tests shipped in one day (an edge-run input the retired regex handled in one pass, invariants the old implementation already satisfied, an entry point whose trimming defused the flagged pattern); review's counterfactual checks caught all three. The same proof discipline already existed piecemeal for moved tests and structural gates — name it once and point to the mechanical proof shapes. Co-Authored-By: Claude <noreply@anthropic.com>
Size Report
Startup median (7 runs, lower is better):
Top changed chunks: no changes in the largest emitted chunks. |
Dropped three bullets: open-before-close (CLI help and device-verification.md own it), don't-remove-without-migration (subsumed by the stronger no-fallback scope rule), and generic Node built-ins advice (engines owns the version). Strengthened the oxfmt rule with the confirmed mechanism: a path argument bypasses ignorePatterns, not just hides drift — one path-scoped run re-quoted 44 excluded conformance corpus files. Co-Authored-By: Claude <noreply@anthropic.com>
|
Second commit extends this into the curation pass requested: audited every bullet against "earned, non-obvious, still true". Three trims — open-before-close (owned by CLI help + device-verification.md, which the routing table already points at), don't-remove-shared-behavior-without-migration (strictly subsumed by the no-fallback/complete-the-migration scope rule), and the generic Node-builtins bullet (engines owns the version; the advice is table stakes). One strengthening from a failure confirmed this very session: Generated by Claude Code |
|
Blocking at Please also restore the three removed AGENTS hard rules. The device-verification document only requires cleanup for manually opened sessions; it does not replace the implementation-level daemon-session-flow invariant. The built-ins preference and shared snapshot/session migration guard likewise are not duplicated by |
This reverts commit 0229cba.
|
Reverted the trim/strengthen commit wholesale in One factual note for the record, offered because this file is exactly about keeping claims true: the Generated by Claude Code |
|
Re-reviewed exact head |
|
Distills this week's review pattern into the guidance docs. Three regression tests shipped vacuous in one day — each passed against the code it claimed to pin (#1541's invariant tests, #1546's edge-run ReDoS case, #1536's trim-defused adversarial input) — and each was caught by the reviewer counterfactual-testing the test, then fixed by the mechanical proof: revert the production change, watch the test fail, quote the number (3,751ms / 3,201ms), restore.
Docs-only.