Research quarantine promotion
Skill Lx050/rubbing-to-knowledge/skills/research-quarantine-promotion
书生国智科探挑战赛 · 赛道六 AI for Social Science · 古代碑帖与拓片 — 可复现证据链科研 Skill 系统:12 个离线、确定性、纯标准库 Skill(哑舍小分队 · 西北大学)
npx -y skills add Lx050/rubbing-to-knowledge --skill research-quarantine-promotionAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
3 things to look at
- 11 days oldThe repository was created 11 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
- no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
- 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
What its author says it does
Copied from the file, not written here
Safely assess one already verified quarantine JSON as bounded inert data, using a preregistered exact binding, an exclusive source-verification receipt, same-buffer identity and strict parsing, content-free structure reports, and offline producer verification. Use when research-source-acquisition has preserved an application/json asset in quarantine and a later internal workflow needs an auditable B0 gate before any IIIF, OCR, schema, literature, or knowledge-graph interpretation; do not use for arbitrary JSON, URL following, semantic validation, canonical materialization, academic admission, or public release.
SKILL.md
10.0 KB, ~2.0k tokens by cl100k_base, as published. Nobody here has run it
Research Quarantine Promotion
Close the trust boundary between downloaded quarantine bytes and a
content-aware research tool. Version 1.0.5 implements only the conservative
strict-json-b0-v1 structure gate: it never copies the payload or emits source
keys, strings, URLs, excerpts, or semantic fields.
Read the contract
Read references/contract.md before authoring a
binding or invoking the CLI. Use these offline schemas for authoring and
review:
promotion-binding-v1.schema.jsonsource-verification-receipt-v1.schema.jsonassessment-report-v1.schema.jsonassessment-event-v1.schema.json
The Python CLI is authoritative. It uses exact-field validation and does not
load a remote schema or follow $ref, @context, URL, file path, or payload
instruction.
Preserve the boundary
- Start only from one successful
research-source-acquisitionrun whose manifest, rights, events, payload identity, and exact file set have been preregistered. - Require a new, exclusive source-verification receipt produced by the pinned
source-acquisition implementation through the binding's dedicated
research-run-ledgerafter the checkpoint. A historical exit code, hand-written JSON, or receipt without matching stdout/output evidence is not a substitute. - Keep the original payload in quarantine. Never move, chmod, rename, delete, overwrite, copy, normalize, or publish it.
- Require binding schema 1.1 identity guards and the metadata-only publish
boundary to pass before any assessment
artifact, including an error artifact, may be created. A receipt, execution
ledger, log directory, or attempt inside the acquisition run—or any overlap
among those four roles—must return without writing. Existing components are
compared by descriptor-backed device/inode ancestry; unresolved components
are conservatively compared after Unicode normalization and case folding.
At command start, the acquisition run, receipt parent, run-ledger parent, and
log directory are opened as one identity-checked descriptor group and kept
alive until preflight returns, assessment publication finishes, or producer
verification returns. Every acquisition metadata, receipt, run-ledger, log,
and payload read is relative to those retained descriptors. Their current
pathnames are rebound to the retained device/inode identities before the first
attempt write, between writes, and after final
fsync; the payload directory entry must also retain the exact source-execution/scanned inode. A late failure removes the still-pinned incomplete attempt instead of leaving pass/block artifacts. - Run
preflightbefore any payload access. Stop unless it reportspayload_opened: false. - Let the binding select the only run, payload identity, profile, provenance material, implementation bytes, receipt path, and output attempt. Do not add ad hoc CLI overrides.
- Treat evidence, rights, receipt, path, hash, race, and ledger failures as
error/invalid, not as JSON content negatives. - Treat only strict JSON syntax or fixed resource-profile failures as
block. - Treat
pass-candidateasstrict-json-structure-check-passed / internal-research-only. It does not establish content truth, IIIF semantics, carrier/locator admission, D2/D3 rights, trusted-human review, L2/L4, ResearchCaseSRC, academic evidence, or public release. - Preserve every pass, block, error, stdout, stderr, report, event chain, and independent review. Never widen the profile and retry under the same promotion ID.
Prepare the binding
Create one strict JSON binding from the schema and contract. Freeze it before the source receipt or purpose-specific B0 payload access. It must bind:
- the preregistration path, SHA-256, and size;
- the acquisition run and all five fixed file identities;
- the future receipt path, its bound parent device/inode, and pinned source verifier;
- a dedicated future run-ledger manifest, its bound parent device/inode, a
pre-created empty
0700artifact directory with bound device/inode, run ID, current Python executable, pinned run-ledger implementation and repo cwd; - the three-layer rights statuses without upgrading them;
- the entire current provenance ledger, head, material registration, and latest verification;
- this script, contract, and four schemas by exact path/hash/size;
- one new attempt directory whose basename is the promotion ID and whose existing parent device/inode is bound;
- the acquisition run's current device/inode, current-user ownership and
exact
0700mode.
Do not derive the profile, rights, output path, limits, or allowed capabilities from the payload. Keep the receipt, run-ledger manifest/log directory and attempt outside the acquisition run and outside one another.
Preflight without opening the payload
python3 scripts/quarantine_promotion.py preflight \
--binding /absolute/promotion-binding.json
Preflight reads only metadata and performs lstat-level payload checks. It may
run before the future receipt exists. Require exit 0, status: ready, and both
payload-open fields to be false. Record the command with
research-run-ledger; a preflight PASS is only permission to complete the
checkpoint and later source verification.
Produce a fresh source receipt
After the preregistration, implementation, tests, independent audit, binding,
preflight evidence, and Git checkpoint are frozen, invoke the pinned source
verifier through the dedicated research-run-ledger command frozen in the
binding. Its child argv is:
python3 ../research-source-acquisition/scripts/source_acquisition.py verify \
--run-directory /absolute/quarantine/acquisition-run \
--run-directory-identity DEVICE:INODE \
--receipt /absolute/new/source-verification-receipt.json \
--receipt-parent-identity DEVICE:INODE
Declare the pinned Python executable, run-ledger script, source-verifier script
and all five acquisition files as exact inputs, and the new receipt as the
only snapshot output. Invoke run-ledger with the binding's manifest-parent and
artifact-directory identity guards, and require its publication_boundaries
evidence to equal the binding. Require research-run-ledger 1.1.3: every
declared input must carry its device/inode identity before and after the child,
and the runner must reserve capacity for the complete start/finish pair before
launch. Promotion reconstructs those identities from current files; for the
payload it uses only descriptor-relative stat, so an identity replacement
fails before payload bytes are opened. Continue only when the two-event ledger
verifies with unchanged input bytes and identities, exit 0, empty stderr,
stdout bytes exactly equal the receipt, ok: true, status: verified, and no
mismatches. Preserve a non-passing receipt as a negative result and stop.
Assess once
python3 scripts/quarantine_promotion.py assess \
--binding /absolute/promotion-binding.json
Assessment verifies the receipt, its dedicated execution ledger and metadata,
opens only the manifest-derived payload relative to the retained bound
acquisition-run descriptor with no-follow/single-link controls, requires that
inode to equal the source-execution identity, hashes and parses the same
bounded buffer with an incremental, no-AST scanner, and keeps all four source
directory pins through a new two-file attempt. The writer pins both the
attempt parent and the newly created
attempt directory through the complete exclusive-write, exact-file-set,
fsync, and failure-cleanup transaction.
The expected attempt-parent identity is carried from the publish boundary into
the descriptor-relative create, so a pathname replacement between validation
and publication fails before mkdir.
Expected exits:
0:pass-candidate;5: contentblock;3: integrityerror;2: invocation error before a safe attempt can be established.
The report contains only identities, gate states, schema-fixed numeric structure counters and limits, capability denials, zero content-exposure counters, and the proof boundary. There is no canonical derivative in B0.
Verify the producer artifact
python3 scripts/quarantine_promotion.py verify \
--attempt-directory /absolute/PROMOTION-ID
Verification reopens the current source and recomputes the complete binding, receipt, metadata, rights, provenance, structure, report, exact two-file set, and event chain while retaining the same four bound directory descriptors. It performs one final pathname and payload-inode rebind immediately before return. It returns the same status class as the assessment when the artifact remains reproducible; any tamper or drift returns integrity exit 3.
Afterward, verify the outer run ledger with --check-files and obtain an
independent no-content-leak/outcome audit.
Stop after B0
Whether the result is pass, block, or error, stop this cycle after producer, run-ledger, and independent verification. Canonical materialization, IIIF field extraction, canvas/frame selection, locator construction, per-frame rights, OCR, evidence-card creation, and knowledge-network work require new Skills or profiles with new preregistration and review. B0 is not an adapter that bypasses those gates.