zrwangems/dsh-zrwang-plugin / ars/skills/sr-screener
sr-screener
Flagged系统综述文献筛选:把方案或 PROSPERO 协议转成资格规则,双盲筛选标题摘要与全文,第三人裁决冲突,输出 PRISMA 2020 计数、EndNote/Zotero RIS 分组与可交付给论文技能的 literature_corpus。Triggers: systematic review screening, screen papers, PRISMA counts, pilot screening, adjudicate conflicts | 文献筛选, 系统综述筛选, PRISMA 计数, 筛选冲突裁决。
_shared/contracts/degradation_registry.json
View on GitHub{
"$comment": "Degradation registry (#511 Part A). Machine-readable INDEX of every graceful-degradation mechanism in ARS: what fails, what degraded state is emitted, where the degradation is visible, who consumes it, and how the terminal-policy layer treats it. The registry INDEXES — it never re-authors: each row's `authority` points at the per-mechanism source of truth (a content anchor that must appear verbatim in the cited file), so semantics live in exactly one place and this file cannot drift into a second authority. Rows use each mechanism's NATIVE semantics (unresolvable / not_checked / PASS WITH NOTES / omit-field / warn-and-continue); there is deliberately NO universal score-cap column — ARS's gate model is tri-state outcomes + advisory suffixes + opt-in terminal policies, not a shared scalar score (#511 'Explicitly rejected: score capping'). Numeric/severity caps appear only where a scale already exists (see compliance_non_sr_warn_cap). Adding a row: fill every required field, cite the authority file with a short verbatim anchor (never a line number — anchors survive drift, line numbers do not), list the tests/lints/schemas that pin the behavior under `pinned_by` (path, or path::test_function), then run scripts/check_degradation_registry.py.",
"registry_version": "1.3.0",
"mechanisms": [
{
"mechanism": "citation_resolver_outage",
"failure_class": "Bibliographic lookup API unreachable (network error, 5xx, rate-limit exhaustion) during citation-existence verification",
"degraded_state": "Per-resolver outcome status 'unreachable' (queried_by null); a TOTAL outage of all applicable resolvers reduces lookup_verified to 'unresolvable' — never to 'true' (an outage is not evidence of existence) and never alone to 'false' (an outage is not fabrication evidence)",
"diagnostic_marker": "resolver_outcomes.<resolver>.status == 'unreachable' in the citation verification summary",
"downstream_consumer": "pipeline_orchestrator_agent finalizer (5-cell gate) and terminal_policies.citation_existence evaluation",
"terminal_policy_effect": "advisory by default; under citation_existence='strict' only lookup_verified=='false' (ID-keyed unmatched, C-V6(a)) refuses — 'unresolvable' from outage is NEVER blocked",
"authority": [
{
"file": "scripts/citation_verification_summary.py",
"anchor": "`unresolvable` otherwise — every applicable resolver `unreachable` (total"
},
{
"file": "scripts/verification_gate/__init__.py",
"anchor": "return _outcome(STATUS_UNREACHABLE, None)"
}
],
"pinned_by": [
"scripts/test_verification_gate.py::test_all_unreachable_is_unresolvable",
"scripts/test_transport_fixture_citation_gate.py::test_total_outage_reduces_to_unresolvable_through_real_clients",
"shared/contracts/passport/citation_verification_summary.schema.json"
]
},
{
"mechanism": "contamination_signal_api_degradation",
"failure_class": "Lookup API (Semantic Scholar / OpenAlex / Crossref / arXiv) degraded while computing a contamination signal at ingest/backfill time",
"degraded_state": "The per-resolver *_unmatched field is OMITTED from contamination_signals — never set to false (which would falsely read 'checked and clean'); the omission is excluded from the k/k_max triangulation count per R-L3-2-C",
"diagnostic_marker": "contamination_signal_omissions.<signal> == 'api_degraded' on the corpus entry (#511 Part A reason-provenance; entries written before this field exists carry no marker — absence of BOTH the signal and the omission reason is indistinguishable from 'not yet computed', which is the pre-#511 live weakness this field closes)",
"downstream_consumer": "pipeline_orchestrator_agent finalizer advisory-suffix matrix (k, k_max) per shared/references/firm_rules.md R-L3-2-C",
"terminal_policy_effect": "advisory by default per R-L3-2-A; a degraded lookup is excluded from the k/k_max count (R-L3-2-C) and which annotation tier fires is determined by the resulting (k, k_max) pair — degradation can therefore change the tier — strengthen it, weaken it, or leave it unchanged — never monotonically weaken it; whether any tier is promoted to a terminal block is decided at the emission boundary by terminal_policies (opt-in strict, contamination_triangulation only), never by the omission itself",
"authority": [
{
"file": "scripts/contamination_signals.py",
"anchor": "None means OMIT the field from the"
},
{
"file": "shared/references/firm_rules.md",
"anchor": "Absent fields are excluded from the count and do not default to either `true` or `false`."
},
{
"file": "shared/references/firm_rules.md",
"anchor": "By default, contamination and temporal-integrity signals never block emission on their own"
}
],
"pinned_by": [
"scripts/test_contamination_signals.py",
"shared/contracts/passport/literature_corpus_entry.schema.json",
"scripts/adapters/tests/test_literature_corpus_entry_schema.py",
"scripts/test_migrate_literature_corpus_to_v3_9_0.py",
"scripts/test_migrate_literature_corpus_to_v3_7_3.py"
]
},
{
"mechanism": "vlm_unavailable",
"failure_class": "No multimodal/vision capability available for VLM figure verification (or figure too simple to warrant it)",
"degraded_state": "VLM verification step is SKIPPED; the Figure Package records 'VLM verification: skipped' instead of a verification section",
"diagnostic_marker": "'VLM verification: skipped' note in the Figure Package output",
"downstream_consumer": "integrity_verification_agent Stage 4.5 Phase C3 figure/table fidelity review",
"terminal_policy_effect": "no terminal policy exists for VLM; a skipped verification can reach PASS WITH NOTES but never converts to FAIL on absence alone (FAIL requires a positive fidelity finding)",
"authority": [
{
"file": "academic-paper/agents/visualization_agent.md",
"anchor": "If not available or figure is simple: skip (note \"VLM verification: skipped\" in Figure Package)"
},
{
"file": "academic-pipeline/agents/integrity_verification_agent.md",
"anchor": "an entry reaches PASS WITH NOTES only when NO FAIL condition is met"
}
],
"pinned_by": []
},
{
"mechanism": "submission_package_incompleteness",
"failure_class": "A submission-package check cannot run (missing artifact, unreadable input, extraction failure) during package verification",
"degraded_state": "Check status 'not_checked' (distinct from pass AND from fail); process exit code 3 = 'passed what was checkable' — distinct from a full pass (0) and from a failure (1)",
"diagnostic_marker": "[NOT-CHECKED] lines + not_checked_count in the verification report header; exit code 3",
"downstream_consumer": "submission-package policy evaluation (advisory vs strict) and the orchestrator invoking scripts/verify_submission_package.py",
"terminal_policy_effect": "advisory: not_checked surfaces but does not refuse (exit 3); under strict, ONLY strict_eligible not_checked checks fail closed (VERIFICATION-INCOMPLETE token, exit 4) — heuristic-classed not_checked stays nonterminal at exit 3 even under strict",
"authority": [
{
"file": "scripts/verify_submission_package.py",
"anchor": "return 3 # \"passed what was checkable\" (§8) — distinct from a full pass"
},
{
"file": "scripts/verify_submission_package.py",
"anchor": "Fail-closed (§5.2): a missing parser/profile must not silently"
},
{
"file": "scripts/verify_submission_package.py",
"anchor": "if c[\"strict_eligible\"] and c[\"status\"] == \"not_checked\")"
}
],
"pinned_by": [
"scripts/test_verify_submission_package.py"
]
},
{
"mechanism": "cross_model_unavailable",
"failure_class": "Transport-level failure of a CONFIGURED cross-model verifier mid-run (API error, rate limit, expired key, selected Codex citation transport unavailable, or a provider-side safety / misalignment intervention that surfaces as an API error or an adapter failure). An unset ARS_CROSS_MODEL is NOT a degradation — it is the standard single-model configuration and produces no error marker or disclosure; an unset or api ARS_CROSS_MODEL_TRANSPORT is likewise ordinary API-route selection, while an invalid selector is a visible configuration error",
"degraded_state": "Warn-and-continue: pipeline proceeds single-model; NEVER blocks on cross-model transport failure; a NOT_SEARCHED grounded-lookup result is NOT a transport failure and does not fall back (it is counted separately and surfaced for re-run/human review)",
"diagnostic_marker": "[CROSS-MODEL-ERROR: reason] log line + required report disclosure 'Cross-model verification was configured but unavailable for this run. Results are single-model only.'",
"downstream_consumer": "integrity_verification_agent (Stage 2.5/4.5 sample checks), devils_advocate reviewers, blind disagreement checkpoints (#518)",
"terminal_policy_effect": "none — cross-model is verification-support, not a gate; its absence is disclosed, never blocking",
"authority": [
{
"file": "shared/cross_model_verification.md",
"anchor": "Continue with single-model verification — never block the pipeline on cross-model failure"
},
{
"file": "shared/cross_model_verification.md",
"anchor": "Cross-model verification was configured but unavailable for this run. Results are single-model only."
},
{
"file": "shared/cross_model_verification.md",
"anchor": "A `NOT_SEARCHED` result is **not** a transport failure and is handled differently."
},
{
"file": "shared/cross_model_verification.md",
"anchor": "An intervention is never a verdict."
}
],
"pinned_by": [
"scripts/check_cross_model_verification_sync.py",
"scripts/test_cross_model_codex_transport.py",
"scripts/check_630_codex_subscription_transport.py"
]
},
{
"mechanism": "compliance_non_sr_warn_cap",
"failure_class": "Compliance checkpoint running outside systematic-review mode (PRISMA-trAIce scope not applicable)",
"degraded_state": "RAISE principles-only evaluation with severity capped at 'warn' — never block (the one registry row with a severity cap, legitimate because compliance already has a native block>warn>info scale; NOT a precedent for score caps elsewhere)",
"diagnostic_marker": "compliance section of the FULL checkpoint dashboard showing principles_only / warn-only evaluation",
"downstream_consumer": "Stage 2.5 / 4.5 integrity gates aggregating compliance decisions via max_severity precedence",
"terminal_policy_effect": "non-SR compliance findings can warn but never block the pipeline; block conditions are SR-mode-only",
"authority": [
{
"file": "shared/compliance_checkpoint_protocol.md",
"anchor": "| Non-SR mode | RAISE principles_only, warn-only (never block) | Same |"
}
],
"pinned_by": [
"scripts/test_check_compliance_report.py"
]
},
{
"mechanism": "re_review_attribution_indeterminate",
"failure_class": "Stage 3' new-issue provenance cannot be established despite the hard-required original manuscript (for example, non-comparable formats or an exact-span comparison that remains unavailable) (#576 §8/SD-9)",
"degraded_state": "Attribution 'indeterminate' — treated as previously_missed for decision purposes (cannot escalate the decision), never silently promoted to regression",
"diagnostic_marker": "[ATTRIBUTION-INDETERMINATE]",
"downstream_consumer": "Stage 3' decision derivation (goalpost guard — indeterminate never appears in Step 2) and integrity_verification_agent at Stage 4.5 (consumes indeterminate records as gate input)",
"terminal_policy_effect": "none — fail-closed on attribution and decision-inert by design; current 1.1 never uses missing original-manuscript bytes as this degradation path",
"authority": [
{
"file": "academic-paper-reviewer/references/re_review_mode_protocol.md",
"anchor": "flagged `[ATTRIBUTION-INDETERMINATE]` — never silently promoted to `regression`"
}
],
"pinned_by": [
"scripts/test_check_re_review_synthesis.py"
]
},
{
"mechanism": "re_review_criteria_layer_absent_no_letter",
"failure_class": "Stage 3' with the Editorial Decision Letter absent from the input manifest: the criterion inheritance chain loses its level-2 (letter Acceptance-criteria) layer (#576 §11)",
"degraded_state": "The §4 chain runs without level 2 — inherited_criterion.letter_text legitimately absent on every item; verification proceeds on the Schema 7 verification_criteria + remaining layers",
"diagnostic_marker": "[CRITERIA-LAYER-ABSENT: no decision letter]",
"downstream_consumer": "Phase 1 pre-commitment records and check_re_review_synthesis.py letter-binding invariants",
"terminal_policy_effect": "none — the letter is not hard-required; the run-level marker plus the existing 'unknown (provenance block absent)' Judge Record rule make the absence visible",
"authority": [
{
"file": "academic-paper-reviewer/references/re_review_mode_protocol.md",
"anchor": "[CRITERIA-LAYER-ABSENT: no decision letter]"
}
],
"pinned_by": [
"scripts/test_check_re_review_synthesis.py::test_letter_fields_absent_when_layer_degraded_passes"
]
},
{
"mechanism": "re_review_criteria_layer_absent_ordinal_mismatch",
"failure_class": "Stage 3' with the letter PRESENT but its Required Item Details block ids not exactly the contiguous sequence R1..Rn against the roadmap's must_fix order — a gap, duplicate, or extra id makes the #576 §5.1 ordinal derivation unsound",
"degraded_state": "The WHOLE letter criterion layer degrades to absent (every item legally omits letter_text/letter_item_ref); the checker RECOMPUTES the contiguity state from the hash-bound letter + roadmap — no carrier field exists",
"diagnostic_marker": "[CRITERIA-LAYER-ABSENT: letter/roadmap ordinal mismatch]",
"downstream_consumer": "Phase 1 pre-commitment records and check_re_review_synthesis.py letter-binding invariants",
"terminal_policy_effect": "none — a malformed letter degrades the layer visibly rather than producing a possibly-wrong join or an abort",
"authority": [
{
"file": "scripts/check_re_review_synthesis.py",
"anchor": "[CRITERIA-LAYER-ABSENT: letter/roadmap ordinal mismatch]"
},
{
"file": "academic-paper-reviewer/templates/editorial_decision_template.md",
"anchor": "Required blocks are exactly `R1..Rn` with no gaps, duplicates, or extras"
}
],
"pinned_by": [
"scripts/test_check_re_review_synthesis.py::test_letter_ordinal_gap_degrades_whole_letter_layer"
]
},
{
"mechanism": "re_review_round1_findings_absent",
"failure_class": "Stage 3' with the Round-1 review findings (Schema 6 reports) absent from the input manifest: the criterion chain's level-3 driving-finding layer has no report-side source (#576 §11)",
"degraded_state": "Level 3 degrades to the Schema 7 TRANSPORTED fields alone (severity/anchor/confidence ride on the roadmap items themselves, #574 A2/A3) — partial coverage survives",
"diagnostic_marker": "[ROUND1-FINDINGS-ABSENT]",
"downstream_consumer": "Phase 1 pre-commitment operationalization and the driving_severity operands of the decision derivation",
"terminal_policy_effect": "none — transported fields keep the derivation sound; a null driving_severity on a legacy item is treated as NOT critical, never fabricated",
"authority": [
{
"file": "academic-paper-reviewer/references/re_review_mode_protocol.md",
"anchor": "[ROUND1-FINDINGS-ABSENT]"
}
],
"pinned_by": []
},
{
"mechanism": "re_review_commitment_evidence_absent",
"failure_class": "Stage 3' with the Response to Reviewers absent: acknowledgment_only commitments (whose evidence IS the letter) cannot be independently confirmed (#576 §11)",
"degraded_state": "Phase 2B runs claim-matching-empty (authors_claim = \"—\"); acknowledgment_only commitments keep their author-declared fulfillment_status UNCHANGED (the closed Schema 11 enum gains no value) but each affected commitment carries the report-level advisory flag; the existing COMMITMENT_GAP rule applies unchanged",
"diagnostic_marker": "[COMMITMENT-EVIDENCE-ABSENT: acknowledgment_only — no response letter]",
"downstream_consumer": "Commitment Ledger Verification output (re-review report) and the revision-loop reader",
"terminal_policy_effect": "none — advisory flag; the P1 authors_claim requirement is satisfied by the letter-absent \"—\" fill (its flag-as-CANNOT_VERIFY consequence is legacy-mode-scoped, #576)",
"authority": [
{
"file": "academic-paper-reviewer/references/re_review_mode_protocol.md",
"anchor": "[COMMITMENT-EVIDENCE-ABSENT: acknowledgment_only — no response letter]"
}
],
"pinned_by": []
},
{
"mechanism": "re_review_routing_degraded_unmapped_labels",
"failure_class": "Stage 3' verifier routing: a routed P1/P2 item's first non-DA label found no card seat, or a non-empty source_reviewer parsed to zero recognized tokens (parse failure) — partial card parsing, even when other cards parsed fine (#576 §10)",
"degraded_state": "The affected items route to EIC; the run-level Routing line is the parameterized degradation form whose payload names every label group and unparsed item — a decision-driving item silently defaulting to EIC cannot hide behind a healthy-looking run",
"diagnostic_marker": "[ROUTING-DEGRADED: unmapped labels — <payload>] (payload grammar: seat-missing `<label> (<item ids>)` groups in label order, then `unparsed <item_id> \"<raw>\"` groups)",
"downstream_consumer": "Judge Record Routing line + Schema 6 judge_record.routing_status (oneOf pattern form — the payload is accountability content, never collapsed to a bare enum)",
"terminal_policy_effect": "none — routing degradation is visible, never a block; card_mapped is reportable ONLY when every routed item's first non-DA label found its seat",
"authority": [
{
"file": "academic-paper-reviewer/references/re_review_mode_protocol.md",
"anchor": "[ROUTING-DEGRADED: unmapped labels — <payload>]"
}
],
"pinned_by": [
"scripts/test_check_re_review_synthesis.py::test_source_reviewer_labels_must_recompute"
]
},
{
"mechanism": "re_review_routing_degraded_cards_unparsable",
"failure_class": "Stage 3' verifier routing: Round-1 cards are present but no non-EIC label parsed anywhere — total card-parse failure (#576 §10)",
"degraded_state": "Routing defaults to EIC-verifies-all; yardstick continuity may still hold (a run can reuse cards for the yardstick AND fail to parse them for routing — the Routing line is orthogonal to Reviewer configuration by design)",
"diagnostic_marker": "[ROUTING-DEGRADED: cards unparsable]",
"downstream_consumer": "Judge Record Routing line + Schema 6 judge_record.routing_status (constant form)",
"terminal_policy_effect": "none — visible total-failure form, never a block",
"authority": [
{
"file": "academic-paper-reviewer/references/re_review_mode_protocol.md",
"anchor": "[ROUTING-DEGRADED: cards unparsable]"
}
],
"pinned_by": []
},
{
"mechanism": "re_review_routing_degraded_no_cards",
"failure_class": "Stage 3' verifier routing with no Round-1 Reviewer Configuration Cards available at all (#576 §10)",
"degraded_state": "The existing [YARDSTICK-REGENERATED] fallback governs configuration; routing defaults to EIC-verifies-all",
"diagnostic_marker": "[ROUTING-DEGRADED: no round-1 cards]",
"downstream_consumer": "Judge Record Routing line + Schema 6 judge_record.routing_status (constant form)",
"terminal_policy_effect": "none — visible; composes with the yardstick-regeneration advisory rather than replacing it",
"authority": [
{
"file": "academic-paper-reviewer/references/re_review_mode_protocol.md",
"anchor": "[ROUTING-DEGRADED: no round-1 cards]"
}
],
"pinned_by": []
},
{
"mechanism": "re_review_legacy_no_contract",
"failure_class": "Stage 3' (or standalone) re-review explicitly run under ARS_RE_REVIEW_LEGACY=1 — the pre-contract single-pass orchestration shape (#576 §12/SD-8)",
"degraded_state": "One-call legacy run, old template; NO traceability sidecar is produced; no byte-equivalence claim is made for LLM prose (the flag preserves ORCHESTRATION shape only). Absent the flag, a run that cannot satisfy the contract prerequisites ABORTS with the specific G0-class reason — never a silent legacy fallback",
"diagnostic_marker": "[LEGACY-NO-CONTRACT] at the top of the Verification Review Report",
"downstream_consumer": "Stage 3' checkpoint reader; integrity_verification_agent at Stage 4.5 (sidecar absence is a visible degradation to note, not a block)",
"terminal_policy_effect": "none — legacy is explicit opt-in; the fail-closed boundary is on the FLAGLESS path (G0-class abort), not on the flagged one",
"authority": [
{
"file": "academic-paper-reviewer/references/re_review_mode_protocol.md",
"anchor": "[LEGACY-NO-CONTRACT]"
},
{
"file": "academic-pipeline/agents/pipeline_orchestrator_agent.md",
"anchor": "Never convert an abort into a decision or a silent legacy run"
}
],
"pinned_by": []
},
{
"mechanism": "re_review_apply_chain_witness_not_run",
"failure_class": "Stage 3' apply-chain witness has no links to check because apply_reports is empty for an all-declined no-op round (#576 current 1.1 §11)",
"degraded_state": "Closed composite state not_run_no_reports (zero reports — never a silent vacuous pass); original_manuscript is hard-required and first_link_not_run belongs only to archived 1.0",
"diagnostic_marker": "apply_chain_witness == 'not_run_no_reports' in DecisionInputs and on the Judge Record Apply-report chain line",
"downstream_consumer": "check_re_review_synthesis.py (witness compared on every emission kind) + Schema 6 judge_record.apply_chain_witness",
"terminal_policy_effect": "none for not_run_no_reports; 'fail' outranks it (fail > not_run_no_reports > pass) and is the G0 manifest_hash_mismatch abort",
"authority": [
{
"file": "scripts/check_re_review_synthesis.py",
"anchor": "fail > not_run_no_reports > pass"
}
],
"pinned_by": [
"scripts/test_check_re_review_synthesis.py::test_no_reports_is_not_run_no_reports_witness",
"scripts/test_check_re_review_synthesis.py::test_recorded_witness_must_match_recomputed"
]
},
{
"mechanism": "write_scope_guard_no_python",
"failure_class": "No real Python interpreter found by the write-scope guard launcher (candidate list exhausted, or only non-functional stubs present, e.g. the 0-byte Microsoft Store `python3` App Execution Alias)",
"degraded_state": "The launcher emits the canonical pass-through hook JSON and exits 0, silent on stderr; the guard is inactive and core skills are unaffected (the guard is optional hardening and ARS core needs no Python). It deliberately emits no permissionDecision — never 'allow' — so the session's normal permission rules still decide, and it never exits non-zero",
"diagnostic_marker": "Silent per call BY DESIGN (per-call stderr on a hot path is exactly the #454 spam the launcher exists to stop); the observable state is hook output equal to the canonical pass-through JSON {\"hookSpecificOutput\":{\"hookEventName\":\"PreToolUse\"}} with no permissionDecision",
"downstream_consumer": "Claude Code PreToolUse permission flow (the pass-through output carries no permissionDecision, so the session's normal permission rules decide)",
"terminal_policy_effect": "none — the write-scope guard sits outside the Material Passport / terminal-policy layer; this failure path resolves to pass-through, never block",
"authority": [
{
"file": "hooks/run_guard.sh",
"anchor": "if no real Python is found, OR the guard subprocess misbehaves, the"
},
{
"file": "hooks/run_guard.sh",
"anchor": "no permissionDecision => falls back to the normal permission"
},
{
"file": "docs/SETUP.md",
"anchor": "if no real Python is found it cleanly no-ops and the guard is simply inactive"
}
],
"pinned_by": [
"scripts/test_run_guard_launcher.py"
]
},
{
"mechanism": "write_scope_guard_no_git_bash",
"failure_class": "Windows host without Git Bash: `hooks.json` invokes the launcher through `bash`, and no POSIX shell is available to run the `.sh` at all",
"degraded_state": "Guard inactive — Claude Code falls back to PowerShell, which cannot run the launcher; this ARS hook never blocks a write (the session's normal permission rules still apply), but the PreToolUse hook logs an error per call (noisy, accepted degradation) until Git Bash is installed",
"diagnostic_marker": "Per-call PreToolUse hook error logged by Claude Code (the launcher never executes on this path, so no ARS-side marker can be emitted)",
"downstream_consumer": "Claude Code PreToolUse permission flow (the hook failure is non-blocking; the session's normal permission rules decide)",
"terminal_policy_effect": "none — the write-scope guard sits outside the Material Passport / terminal-policy layer; on this path the hook simply does not run, so it can block nothing",
"authority": [
{
"file": "hooks/run_guard.sh",
"anchor": "with no Git Bash, CC falls back to PowerShell which can't"
},
{
"file": "hooks/run_guard.sh",
"anchor": "the guard is then inactive (accepted degradation"
},
{
"file": "docs/SETUP.md",
"anchor": "the guard is inactive and the `PreToolUse` hook will log an error per call rather than no-op quietly"
}
],
"pinned_by": []
},
{
"mechanism": "write_scope_guard_no_timeout_binary",
"failure_class": "No usable `timeout` binary on PATH when the launcher wall-clock-bounds its interpreter probes and the guard subprocess: none present, or one that rejects the GNU `timeout <N>s cmd` form (the Windows built-in timeout.exe under Git for Windows, #970)",
"degraded_state": "run_bounded switches to the portable background-and-watchdog fallback: same wall-clock bound (ARS_PROBE_BOUND, integer seconds, default 3) and same pass-through-on-overrun posture — a bounding-mechanism swap, never a change to the security decision",
"diagnostic_marker": "None when healthy (the two bounding paths are decision-equivalent); an overrun on either path surfaces only as the canonical pass-through output in place of a guard decision",
"downstream_consumer": "Claude Code PreToolUse permission flow (the pass-through output carries no permissionDecision, so the session's normal permission rules decide)",
"terminal_policy_effect": "none — the write-scope guard sits outside the Material Passport / terminal-policy layer; the healthy fallback still forwards the guard's real decision (including deny) — only an overrun resolves to pass-through, never block",
"authority": [
{
"file": "hooks/run_guard.sh",
"anchor": "No usable `timeout` binary: a portable background-and-watchdog fallback"
},
{
"file": "hooks/run_guard.sh",
"anchor": "never the security decision"
}
],
"pinned_by": [
"scripts/test_run_guard_launcher.py"
]
},
{
"mechanism": "write_scope_guard_subprocess_misbehaves",
"failure_class": "Guard subprocess misbehaves under a healthy launcher — non-zero exit, wall-clock overrun, or output failing the launcher's validity check (anything that is not a JSON object carrying a top-level hookSpecificOutput key; the check is deliberately this shallow and does not validate the full hook schema) — or a missing guard script, or the launcher itself fails internally (mktemp failure, SELF_DIR self-resolution failure)",
"degraded_state": "The launcher emits the canonical pass-through hook JSON and exits 0; the guard's stderr is dropped on this path (relayed only on the healthy path) so a broken guard cannot spam the hot path, and a mktemp failure degrades to pass-through rather than falling back to a predictable (symlink-attackable) temp path",
"diagnostic_marker": "Canonical pass-through JSON in place of a guard decision; the guard's own stderr advisories are absent",
"downstream_consumer": "Claude Code PreToolUse permission flow (the pass-through output carries no permissionDecision, so the session's normal permission rules decide)",
"terminal_policy_effect": "none — the write-scope guard sits outside the Material Passport / terminal-policy layer; this failure path resolves to pass-through, never block",
"authority": [
{
"file": "hooks/run_guard.sh",
"anchor": "Guard broke / timed out / produced invalid output. Degrade, don't block, don't spam."
},
{
"file": "hooks/run_guard.sh",
"anchor": "degrade safely to pass-through instead of writing a guessable path"
},
{
"file": "hooks/run_guard.sh",
"anchor": "SELF_DIR=$(CDPATH= cd -- \"$(dirname -- \"$0\")\" && pwd) || emit_passthrough_and_exit"
}
],
"pinned_by": [
"scripts/test_run_guard_launcher.py"
]
},
{
"mechanism": "write_scope_guard_payload_capacity_edge",
"failure_class": "A POSIX shell caps the payload variable the launcher holds and replays on a multi-megabyte Write payload (the launcher buffers stdin in a shell variable by design)",
"degraded_state": "No pinned outcome — the launcher documents this as an accepted, unfixed narrow case and neither enforces nor tests a specific behavior for it. The row exists for inventory completeness; no pass-through or other deterministic resolution is claimed",
"diagnostic_marker": "None defined for this edge (no code path detects a capped payload)",
"downstream_consumer": "Claude Code PreToolUse permission flow (whatever the launcher emits in this case is consumed there; no specific emission is claimed)",
"terminal_policy_effect": "none — the write-scope guard sits outside the Material Passport / terminal-policy layer; the launcher's design intent is non-blocking, but for this edge no enforced outcome is claimed",
"authority": [
{
"file": "hooks/run_guard.sh",
"anchor": "documented degradation rather than fixed speculatively"
},
{
"file": "hooks/run_guard.sh",
"anchor": "A multi-megabyte Write payload on a POSIX shell that caps variable"
}
],
"pinned_by": []
},
{
"mechanism": "acronym_check_not_checked",
"failure_class": "The acronym check cannot read its input (not a Markdown or plain-text file, not UTF-8 text, an unknown scope name, an unreadable allowlist file, or no requested scope with a line of prose in the input, as in an empty file), or the script cannot run at all (for example, no Python interpreter)",
"degraded_state": "not_checked: the script exits 2 and its report states the reason; no finding is emitted. When the script cannot run, the caller reports that the acronym check did not run",
"diagnostic_marker": "JSON status not_checked with a closed reason code; Markdown report line 'Not checked: <reason>.' ('未檢查:<原因>。' in zh-TW)",
"downstream_consumer": "The caller that runs scripts/check_acronyms.py, the writer or abstract_bilingual_agent that receives the report, the Editorial Decision Letter that carries it as an advisory attachment after a review, and the user, who is shown it",
"terminal_policy_effect": "none: the acronym check is advisory; findings and degraded states never block a handoff, change a decision, or stop a run",
"authority": [
{
"file": "scripts/check_acronyms.py",
"anchor": "Exit 2: nothing was checked (``not_checked``: unsupported format, unreadable"
},
{
"file": "academic-paper/references/writing_quality_check.md",
"anchor": "If the script cannot run at all, say that the acronym check did not run; never report it as clean."
}
],
"pinned_by": [
"scripts/test_check_acronyms.py::test_clean_result_is_distinct_from_not_checked",
"scripts/test_check_acronyms.py::test_cli_exit_codes_and_outputs"
]
},
{
"mechanism": "acronym_check_partial_coverage",
"failure_class": "The acronym check reads the input only in part: an acronym uses a definition form the check does not read, or a parenthetical the check cannot confirm as its definition; an abstract is in a language the check does not read; or a requested scope is not in the input",
"degraded_state": "partial: the script exits 0, its coverage line says partial, and the report lists each unread part; an acronym with an unread or unconfirmed definition gets no finding in that scope",
"diagnostic_marker": "JSON status partial, with coverage marking a requested scope not_in_input, or a non-empty coverage_limits or unread_sections; Markdown report coverage line ending '(partial)', a 'Not checked:' line listing coverage_limits and unread_sections when either is non-empty, and a 'Not in this input:' line when a requested scope is absent ('(部分)', '未檢查:', and '此檔沒有:' in zh-TW)",
"downstream_consumer": "The caller that runs scripts/check_acronyms.py, the writer or abstract_bilingual_agent that receives the report, the Editorial Decision Letter that carries it as an advisory attachment after a review, and the user, who is shown it",
"terminal_policy_effect": "none: the acronym check is advisory; a partial report is not a clean result and blocks nothing",
"authority": [
{
"file": "scripts/check_acronyms.py",
"anchor": "status = \"partial\" if limits or doc.unread_sections or missing else \"checked\""
},
{
"file": "academic-paper/references/writing_quality_check.md",
"anchor": "A `partial` report lists what it did not read"
}
],
"pinned_by": [
"scripts/test_check_acronyms.py::test_unread_definition_form_is_a_coverage_limit_not_a_finding",
"scripts/test_check_acronyms.py::test_a_parenthetical_it_cannot_confirm_is_a_coverage_limit",
"scripts/test_check_acronyms.py::test_unread_abstract_section_makes_coverage_partial",
"scripts/test_check_acronyms.py::test_requested_scope_absent_from_the_input"
]
},
{
"mechanism": "panel_reference_check_not_checked",
"failure_class": "The suggested-reference check can read none of the reviewer report files it is given (missing, unreadable, or not UTF-8 text), or the script cannot run at all (for example, no Python interpreter)",
"degraded_state": "not_checked: the script exits 2 and its report states the reason; no reference is looked up. When the script cannot run, the dispatching session writes one line saying the suggested-reference check did not run",
"diagnostic_marker": "JSON status not_checked with reason no_readable_report; Markdown report line 'Not checked: no reviewer report file could be read.' ('未檢查:沒有任何一份審稿報告檔可以讀取。' in zh-TW)",
"downstream_consumer": "The dispatching session that runs scripts/verify_panel_recommendations.py after a review, the Editorial Decision Letter that carries the report as an advisory attachment, and the user, who is shown it",
"terminal_policy_effect": "none: the suggested-reference check is advisory; findings and degraded states never change a decision, a roadmap entry, or a re-review verdict",
"authority": [
{
"file": "scripts/verify_panel_recommendations.py",
"anchor": "return 2 if report[\"status\"] == \"not_checked\" else 0"
},
{
"file": "academic-paper-reviewer/SKILL.md",
"anchor": "When neither seat ran, or the script cannot run, the section is one line saying the suggested-reference check did not run."
}
],
"pinned_by": [
"scripts/test_verify_panel_recommendations.py::test_unreadable_reports_are_not_checked",
"scripts/test_verify_panel_recommendations.py::test_cli_exit_codes_and_json"
]
},
{
"mechanism": "panel_reference_check_partial",
"failure_class": "The suggested-reference check reads the panel's suggestions only in part: a report file cannot be read while another can, a seat's report has no yaml suggested-references block, a block is empty, not valid YAML, or not a list, an item is not a key-value entry or has no title, or an item's result is no_answer (no index matches, none reports no record for its DOI or arXiv ID, and either the item has such an identifier or no index answered for its title) or its lookup fails",
"degraded_state": "partial: the script exits 0, its coverage line says partial, and the report lists each part it did not read; an item whose deciding indexes did not answer is shown as not checked, never as not found; with no item read, the report says no suggested reference could be read, never that no seat recommends one",
"diagnostic_marker": "JSON status partial, with a non-empty problems list on a report record, a non-empty item_problems list, or an item whose result is no_answer; Markdown report coverage line ending '(partial).', with a 'Not read:' list, an item whose result is 'Not checked: an index that could settle it did not answer', or both ('(部分)。', '未讀取:', and '未檢查:能判定此筆的索引沒有回應' in zh-TW)",
"downstream_consumer": "The dispatching session that runs scripts/verify_panel_recommendations.py after a review, the Editorial Decision Letter that carries the report as an advisory attachment, and the user, who is shown it",
"terminal_policy_effect": "none: the suggested-reference check is advisory; a partial report is not a clean result and blocks nothing",
"authority": [
{
"file": "scripts/verify_panel_recommendations.py",
"anchor": "report[\"status\"] = \"partial\" if incomplete else \"checked\""
}
],
"pinned_by": [
"scripts/test_verify_panel_recommendations.py::test_missing_block_and_bad_yaml_make_the_report_partial",
"scripts/test_verify_panel_recommendations.py::test_one_unreadable_report_of_two_is_partial",
"scripts/test_verify_panel_recommendations.py::test_no_index_answering_is_no_answer_and_partial"
]
},
{
"mechanism": "panel_claim_check_not_run",
"failure_class": "The reviewer claim check cannot complete: no manuscript file, the judge subagent does not return, the key or the reply cannot be read, or the manuscript differs from the one the prompt was built from",
"degraded_state": "not run: the script exits 2 with a STOP line, and the Editorial Decision Letter attachment is one line saying the reviewer claim check did not run",
"diagnostic_marker": "stderr line starting 'STOP: ' and exit code 2; the one-line attachment under '## Attachment: Reviewer Claim Check (advisory, #1002)'",
"downstream_consumer": "The dispatching session that writes the attachment, the reader of the Editorial Decision Letter, and the user, who is shown it",
"terminal_policy_effect": "none: the reviewer claim check is advisory; a check that did not run is not a clean result and blocks nothing",
"authority": [
{
"file": "scripts/audit_panel_claims.py",
"anchor": "raise Stop(\"the manuscript is not the one the prompt was built from\")"
},
{
"file": "academic-paper-reviewer/SKILL.md",
"anchor": "the section is one line saying the reviewer claim check did not run"
}
],
"pinned_by": [
"scripts/test_audit_panel_claims.py::test_a_changed_manuscript_stops_the_render",
"scripts/test_audit_panel_claims.py::test_the_cli_refuses_an_unknown_seat_and_a_repeated_seat",
"scripts/test_audit_panel_claims.py::test_a_malformed_key_stops_with_exit_2",
"scripts/test_audit_panel_claims.py::test_an_unwritable_out_dir_stops_with_exit_2"
]
},
{
"mechanism": "panel_claim_check_partial",
"failure_class": "The reviewer claim check reads the panel only in part: a seat's report cannot be read, or the judge's reply gives no usable verdict for some findings (none, an invalid one, or more than one)",
"degraded_state": "partial: the script exits 0, and the report names each unread seat and counts the findings with no usable verdict; those findings are neither contradicted nor cleared",
"diagnostic_marker": "Markdown report line 'Coverage: partial. Not read: <seats>.' or 'Coverage: partial. <n> findings have no usable verdict.' ('涵蓋範圍:部分。' in zh-TW); with no report read, 'No report could be read, so no finding was checked.'",
"downstream_consumer": "The reader of the Editorial Decision Letter that carries the report as an advisory attachment, and the user, who is shown it",
"terminal_policy_effect": "none: the reviewer claim check is advisory; a partial report is not a clean result and blocks nothing",
"authority": [
{
"file": "scripts/audit_panel_claims.py",
"anchor": "if report[\"answered\"] < report[\"findings\"]:"
}
],
"pinned_by": [
"scripts/test_audit_panel_claims.py::test_missing_verdicts_make_the_coverage_partial",
"scripts/test_audit_panel_claims.py::test_an_unreadable_report_is_listed",
"scripts/test_audit_panel_claims.py::test_an_id_answered_twice_has_no_usable_verdict",
"scripts/test_audit_panel_claims.py::test_an_unread_seat_makes_the_coverage_partial_and_none_read_is_not_none"
]
}
]
}
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 |