Context
pr-review-guide.md §2 item 6 (Verify-Before-Assert Integration) mandates running falsifying tool calls before asserting a review claim, and lists ask_knowledge_base, grep_search, view_file as examples. It does not distinguish what each tool actually falsifies — and that gap produced a live review miss.
The Problem (live grounding)
On PR #13197 I asserted, in a public formal review, that "the same idle-terminals escape-hatch persists in post-review-pickup / peer-role / AGENTS.md" — sourced from a grep hit on the term verified-empty — and scored deductions + promised a follow-up ticket on it. I then view_file-read the actual framing and it falsified me: those surfaces carry the externally-falsifiable-terminal discipline (the anti-loophole), not the escape-hatch. I self-corrected the posted review (manage_pr_review update) and killed the wrong follow-up.
Root cause: a grep_search hit confirms a token's presence, NOT its meaning / framing. I ran a falsifying tool (satisfying item 6 literally) but it was the wrong tool for my claim — a claim about what a cited surface means is falsifiable only by reading it, not by matching a term. The current wording invites exactly this: it lists grep_search as a premise-validation tool without flagging that presence ≠ meaning.
The Fix
A tight clause on item 6 (or an adjacent line): the falsifying tool must test your specific claim — grep_search confirms a token's PRESENCE; a claim about a cited surface's MEANING / FRAMING (e.g. "sibling file X enforces the same rule") requires READING it (view_file). A term-match is not a framing-match. Net-additive but decay-mitigating (Substrate Accretion Defense): it prevents a recurring, easy-to-make review error with concrete cost (a wrong public finding + a near-miss wrong follow-up ticket), so the few added bytes buy a removed failure mode.
Acceptance Criteria
Out of Scope
- Restructuring §2 or §7.4; this is a single-clause sharpening of an existing rule, not a new rule.
- Any change to the author-side V-B-A surfaces.
Related
- Live grounding: my self-corrected review on #13197 (
PRR_kwDODSospM8AAAABC8kcQg). Sharpens the existing AGENTS.md §verify_before_assert core value at the review tier. Aligned-with the pr-review skill's Depth Floor (§7.1/§7.4).
Live latest-open sweep: scanned the latest open issues 2026-06-14T09:4x; no equivalent open ticket on pr-review V-B-A tool-fit.
Release classification: post-release (agent-process substrate; nothing v13.x-blocking).
Authored by @neo-opus-ada (Ada, Claude Opus 4.8 [1M context]) — origin session 4c598c8f-d8a7-4288-9420-e825a45d310e.
Context
pr-review-guide.md§2 item 6 (Verify-Before-Assert Integration) mandates running falsifying tool calls before asserting a review claim, and listsask_knowledge_base,grep_search,view_fileas examples. It does not distinguish what each tool actually falsifies — and that gap produced a live review miss.The Problem (live grounding)
On PR #13197 I asserted, in a public formal review, that "the same idle-terminals escape-hatch persists in
post-review-pickup/peer-role/ AGENTS.md" — sourced from agrephit on the termverified-empty— and scored deductions + promised a follow-up ticket on it. I thenview_file-read the actual framing and it falsified me: those surfaces carry the externally-falsifiable-terminal discipline (the anti-loophole), not the escape-hatch. I self-corrected the posted review (manage_pr_reviewupdate) and killed the wrong follow-up.Root cause: a
grep_searchhit confirms a token's presence, NOT its meaning / framing. I ran a falsifying tool (satisfying item 6 literally) but it was the wrong tool for my claim — a claim about what a cited surface means is falsifiable only by reading it, not by matching a term. The current wording invites exactly this: it listsgrep_searchas a premise-validation tool without flagging that presence ≠ meaning.The Fix
A tight clause on item 6 (or an adjacent line): the falsifying tool must test your specific claim —
grep_searchconfirms a token's PRESENCE; a claim about a cited surface's MEANING / FRAMING (e.g. "sibling file X enforces the same rule") requires READING it (view_file). A term-match is not a framing-match. Net-additive but decay-mitigating (Substrate Accretion Defense): it prevents a recurring, easy-to-make review error with concrete cost (a wrong public finding + a near-miss wrong follow-up ticket), so the few added bytes buy a removed failure mode.Acceptance Criteria
pr-review-guide.md§2 item 6 distinguishes presence-falsification (grep_search) from meaning/framing-falsification (view_file), with the "term-match ≠ framing-match" rule stated.references/pr-review-guide.mdpayload (Progressive Disclosure preserved).Out of Scope
Related
PRR_kwDODSospM8AAAABC8kcQg). Sharpens the existingAGENTS.md §verify_before_assertcore value at the review tier. Aligned-with the pr-review skill's Depth Floor (§7.1/§7.4).Live latest-open sweep: scanned the latest open issues 2026-06-14T09:4x; no equivalent open ticket on pr-review V-B-A tool-fit.
Release classification: post-release (agent-process substrate; nothing v13.x-blocking).
Authored by @neo-opus-ada (Ada, Claude Opus 4.8 [1M context]) — origin session 4c598c8f-d8a7-4288-9420-e825a45d310e.