deweight-human-anchor-bind — independently scanned and version-tracked by SaferSkills.
SaferSkills independently audited deweight-human-anchor-bind (Agent Skill) and scored it 100/100 (green). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 0 high-severity and 0 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.
Findings & checks · 0 flagged
Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
防止模型从零凭感觉改写。每个待改段落必须先绑定真人写作实践卡。
../../_shared/references/human-writing-practice-cards/00_A1-A39_实践卡矩阵.md../../_shared/references/human-writing-practice-cards/00_写作功能_实践卡矩阵.md../../_shared/references/human-writing-practice-cards/*/*.md../../_shared/references/rewrite-prescription-rubric.md../../_shared/references/real-example-index.md../../_shared/references/anchor-db-schema.md../../_shared/references/anchor-curation-review-gate.md../../_shared/assets/academic_deweight_anchors.sqlite../../_shared/scripts/query_anchor_db.py../../_shared/scripts/review_anchor_candidates.py../../_shared/references/templates/rewrite-prescription-template.md../../_shared/scripts/check_rewrite_prescription.pydeweight-intake-anchor.deweight-coverage-scan.For each suspect paragraph:
paragraph_id:
semantic_anchor:
writing_function:
hit_modes:
human_anchor:
a_code_cards:
function_cards:
paper_practice_cards:
source_anchor_origin:
human_anchor_evidence:
quote_exact:
source_locator:
source_decision:
observed_human_move:
why_this_quote_fits_user_paragraph:
transferable_structure:
non_transferable_surface:
adaptation_plan:
rewrite_operation:
delete:
keep:
restructure_as:
evidence_boundary:
sentence_rhythm:
citation_number_policy:
forbidden:
- no new fact
- no claim upgrade
- no citation or number drift
- no OCR/MinerU noise copying
verification_focus:A valid practice card must include:
sourcelocationlanguagewriting_functionquote or quote_shortstructure or structure_skeletonanti_ai_modesreuse_rule or rewrite_recipeboundaryRuntime cards currently use the real-paper card fields documented in real-example-index.md; do not require nonexistent fields if an equivalent field is present.
Portable DB entries with source_slug beginning corpus- are mechanically extracted candidates until a curator reviews them. Before binding one of these entries as human_anchor, check anchor-curation-review-gate.md and require an anchor_review_decisions row with:
decision: keepIf the candidate is unreviewed, pending, rejected, or marked needs_source_check, return:
blocked: unreviewed_human_anchor
paragraph_id:
candidate_anchor:
required_action: curator review must mark decision=keep before rewrite bindingnegative_candidate rows are counterexamples only. They preserve ai_judgement="no, style-only" and must never be used as positive practice cards.
Binding must put the real human source wording into the prescription context. Do not bind from an abstract label alone.
For every bound suspect paragraph, include:
quote_exact: the original human source excerpt that the model should read before rewriting.source_locator: file/card/DB locator for the quote.source_decision: runtime_card, curated_anchor, or keep for reviewed corpus anchors.observed_human_move: what the quoted author actually does in the paragraph.why_this_quote_fits_user_paragraph: why the user's paragraph has the same writing problem.transferable_structure: the writing move to imitate.non_transferable_surface: words, entities, data, examples, and topic-specific phrasing that must not be copied.adaptation_plan: how to move from the user's facts to a rewritten paragraph.This is not optional analysis. It is the context bridge that prevents "A-code template rewriting". If no real quote can be shown, return:
blocked: missing_human_quote_evidence
paragraph_id:
needed_quote:If no suitable card exists, return:
blocked: missing_human_anchor
paragraph_id:
needed_anchor:Do not invent a card during rewrite.
Before handoff to rewrite, the prescription should pass:
python3 ../../_shared/scripts/check_rewrite_prescription.py <rewrite-prescription.md>For newly ingested corpus batches, initialize review rows before using candidates:
python3 ../../_shared/scripts/review_anchor_candidates.py ../../_shared/assets/academic_deweight_anchors.sqlite --sync-pending --batch corpus-20260505 --summary~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.