zrwangems/dsh-zrwang-plugin / ars/skills/deep-research
deep-research
Flagged学术研究前置:把模糊想法收敛成可研究的问题,并做文献综述、事实核查、系统综述(PRISMA)与三向文献扫描。Triggers: research, deep research, literature review, systematic review, meta-analysis, PRISMA, fact-check, guide my research | 研究, 文献综述, 系统性文献回顾, 事实核查, 引导我研究, 帮我确定研究问题。输出 RQ Brief、方法论蓝图、带核验的注释书目。
_shared/contracts/passport/claim_audit_result.schema.json
View on GitHub{
"$schema": "https://json-schema.org/draft/2020-12/schema",
"$id": "https://github.com/Imbad0202/academic-research-skills/shared/contracts/passport/claim_audit_result.schema.json",
"title": "Material Passport Claim Audit Result Entry",
"description": "One entry in Material Passport Schema 9 claim_audit_results[] aggregate. Per-claim audit result emitted by claim_ref_alignment_audit_agent (v3.8.0) after the Cite-Time Provenance Finalizer pass at the Stage 4 -> Stage 5 boundary. Cross-field invariants INV-1..INV-19 are lint-enforced in scripts/check_claim_audit_consistency.py — JSON Schema cannot express the conditional matrix relating (judgment, audit_status, defect_stage, ref_retrieval_method). Three sibling entry-types ride in their own aggregates because they cannot satisfy the required-field shape of this schema: uncited_assertion (no ref_slug; §3.3), claim_drift (no judge invocation; §3.4), constraint_violation (uncited HIGH-WARN; §3.5). See docs/design/2026-05-15-issue-103-claim-alignment-audit-spec.md §3.1 + §5 for full rationale.",
"type": "object",
"additionalProperties": false,
"required": [
"claim_id",
"scoped_manifest_id",
"claim_text",
"ref_slug",
"anchor_kind",
"anchor_value",
"judgment",
"audit_status",
"defect_stage",
"rationale",
"judge_model",
"judge_run_at",
"ref_retrieval_method"
],
"properties": {
"claim_id": {
"type": "string",
"pattern": "^C-[0-9]{3,}$",
"description": "Per-claim identifier scoped by scoped_manifest_id. The pair (scoped_manifest_id, claim_id) is the joinable key; bare claim_id alone collides across manifests in the same passport (M-INV-1 permits cross-manifest C-001 reuse)."
},
"scoped_manifest_id": {
"type": "string",
"pattern": "^M-[0-9]{4}-[0-9]{2}-[0-9]{2}T[0-9]{2}:[0-9]{2}:[0-9]{2}Z-[0-9a-f]{4}$",
"description": "Points to the owning claim_intent_manifest.manifest_id. INV-15 enforces cross-array integrity. MANIFEST-MISSING fallback uses the sentinel value M-0000-00-00T00:00:00Z-0000."
},
"claim_text": { "type": "string", "minLength": 1, "maxLength": 2000 },
"ref_slug": {
"type": "string",
"minLength": 1,
"description": "Slug from the v3.7.3 <!--ref:slug--> marker the audited citation carries. Uncited claims do NOT emit into this aggregate; they ride in uncited_assertions[] (§3.3) or constraint_violations[] (§3.5)."
},
"anchor_kind": {
"enum": ["quote", "page", "section", "paragraph", "none"],
"description": "Carried from the v3.7.3 <!--anchor:<kind>:<value>--> marker. anchor_kind=none triggers INV-6 (firm-rule defense-in-depth surface)."
},
"anchor_value": {
"type": "string",
"description": "URL-decoded anchor value. When anchor_kind=none, MUST be the empty sentinel string (INV-6). When anchor_kind != none, INV-16 forbids empty values after whitespace strip."
},
"judgment": {
"enum": ["SUPPORTED", "UNSUPPORTED", "AMBIGUOUS", "RETRIEVAL_FAILED"],
"description": "Judge verdict per the §4 step 5 prompt template. RETRIEVAL_FAILED is NOT a judge label — it is set when retrieval / anchorless / tool-failure prevents the judge from running (per error-handling §4 step 9)."
},
"audit_status": {
"enum": ["completed", "inconclusive"],
"description": "completed = judge ran (or RETRIEVAL_FAILED with deterministic verdict like retrieval_existence). inconclusive = judge could not run and no deterministic verdict was reached."
},
"defect_stage": {
"comment": "Three categories of finding are intentionally NOT defect_stages of claim_audit_result and use their own §3.3/§3.4/§3.5 entry-type schemas: (a) uncited_assertion findings — no ref_slug to evaluate, see uncited_assertion.schema.json (§3.3); (b) claim_intent drift findings — detected by manifest set-diff not by judge, see claim_drift.schema.json (§3.4) (drift never produces a claim_audit_result row because the judge is never invoked for these — the only signal is intended ≠ emitted manifest-set difference); (c) uncited constraint violations — claim has no ref_slug AND violates an MNC/NC rule, see constraint_violation.schema.json (§3.5) (HIGH-WARN gate-refuse but no ref to bind into a claim_audit_result row). 8 values listed below.",
"enum": [
"retrieval_existence",
"metadata",
"source_description",
"citation_anchor",
"synthesis_overclaim",
"negative_constraint_violation",
"not_applicable",
null
]
},
"rationale": {
"type": "string",
"minLength": 1,
"maxLength": 2000,
"description": "One-sentence judge rationale, OR a fault-class-tagged audit-infrastructure failure string for ref_retrieval_method=audit_tool_failure (INV-14), OR the canonical 'v3.7.3 R-L3-1-A violation ...' prefix for anchor_kind=none (INV-6)."
},
"judge_model": { "type": "string", "minLength": 1 },
"judge_run_at": { "type": "string", "format": "date-time" },
"ref_retrieval_method": {
"enum": ["api", "manual_pdf", "failed", "not_attempted", "not_found", "audit_tool_failure"],
"description": "Retrieval pathway. failed = paywall (permanent license restriction, LOW-WARN). not_attempted = anchor=none short-circuit (INV-11). not_found = fabricated reference (INV-12, HIGH-WARN). audit_tool_failure = transient infrastructure outage (INV-14, MED-WARN). See §4 step 9 error handling for the failed/audit_tool_failure permanence discriminator."
},
"upstream_owner_agent": {
"enum": [
"synthesis_agent",
"draft_writer_agent",
"report_compiler_agent",
null
],
"description": "Prose-producing agent that owns the audited claim. Mirrors emitted_by in claim_intent_manifest.schema.json; bibliography_agent is excluded (it never emits claim_audit_results[] entries)."
},
"violated_constraint_id": {
"type": ["string", "null"],
"pattern": "^(NC-C[0-9]{3,}-[0-9]+|MNC-[0-9]+)$",
"description": "Required when defect_stage=negative_constraint_violation (INV-7). Null otherwise. Canonical parse rule per INV-17: NC-C{n}-{m} encodes the parent claim_id digits as {n} (NO inner hyphen between C and {n})."
},
"upstream_dispute": {
"type": ["string", "null"],
"maxLength": 1000,
"description": "Free-form dispute log entry. INV-9: meaningful only when defect_stage is a substantive defect (not null, not not_applicable)."
},
"audit_run_id": {
"type": "string",
"pattern": "^[0-9]{4}-[0-9]{2}-[0-9]{2}T[0-9]{2}:[0-9]{2}:[0-9]{2}Z-[0-9a-f]{4}$",
"description": "Identifier for the audit run that produced this entry. Excluded from the cache key (see §4 step 3) so different audit runs over the same (claim, ref, anchor, source, constraint-set, judge_model) tuple share the cached verdict."
},
"sub_claim_breakdown": {
"type": "array",
"minItems": 2,
"description": "Per-sub-claim decomposition (issue #213). Present iff the judge emitted prompt-verdict PARTIAL: >=2 sub-claims with >=1 SUPPORTED and >=1 non-SUPPORTED, normalized to judgment=UNSUPPORTED + defect_stage=source_description. Its presence IS the machine-readable partial-support signal (consumers read this, not defect_stage). Absent on single-claim and fully-unsupported verdicts. The cross-field shape (true-partial + judgment + defect_stage pinning) is INV-19, lint-enforced in scripts/check_claim_audit_consistency.py.",
"items": {
"type": "object",
"additionalProperties": false,
"required": ["sub_claim_text", "sub_verdict"],
"properties": {
"sub_claim_text": { "type": "string", "minLength": 1, "maxLength": 1000 },
"sub_verdict": { "enum": ["SUPPORTED", "UNSUPPORTED", "AMBIGUOUS"] },
"evidence_pointer": { "type": ["string", "null"], "maxLength": 1000 }
}
}
}
}
}
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| add-product-e2e-evalpaperclipai/paperclipAdd or extend a Paperclip full-stack runner E2E workflow, fixture, matcher, or report evidence path for local or Daytona execution. | Stars: 115k | Updated | Claude CodeCodex | Verified | Install |
| add-runner-evalpaperclipai/paperclipAdd or extend a Paperclip Runner protocol evaluation definition, roster, assertion, or report fixture with provenance and narrow validation. | Stars: 115k | Updated | Claude CodeCodex | Verified | Install |
| diagnose-why-work-stoppedpaperclipai/paperclipCompatibility entry point for failed, stalled, looping, or over-recovered Paperclip work. Use fix-failed-paperclip-run for the full diagnosis, regression, repair, real verification, and green-PR workflow; honor an explicitly requested diagnosis-only scope. | Stars: 115k | Updated | Claude CodeCodex | Verified | Install |
| fix-failed-paperclip-runpaperclipai/paperclipDiagnose and fix failed, stuck, slow, or incorrectly blocked Paperclip tasks and runs. Use with a task/run URL or failure report to trace the root cause, prove a regression red then green, check second-order effects, verify the real workflow, and deliver a PR with passing checks. | Stars: 115k | Updated | Claude CodeCodex | Verified | Install |