@tiangong-ai/cli 0.0.35 → 0.0.37

package/AGENTS.md

@@ -11,12 +11,14 @@ checkPaths:
   - AGENTS.md
   - README.md
   - package.json
+  - .dockerignore
+  - Dockerfile.clean-test
   - .github/workflows/**
   - .docpact/config.yaml
   - docs/agents/**
   - src/**
-lastReviewedAt: 2026-08-13
-lastReviewedCommit: df263405bccb21cd5838d92d82dccf80ca01acb1
+lastReviewedAt: 2026-08-16
+lastReviewedCommit: 622103bad4c33f09cfadda86d68f8ef113fa6804
 ---
 
 # Tiangong AI CLI Contract
@@ -62,7 +64,7 @@ This repository owns the Tiangong AI command-line interface.
 Run before delivery:
 
 ```bash
-npm run test:clean
+npm run test:clean:cold
 npm run lint
 npm run build
 npm test
@@ -71,9 +73,17 @@ docpact validate-config --root . --strict
 docpact lint --root . --worktree --mode enforce
 ```
 
-For Auto Research changes, `npm run test:clean` is the authoritative TDD gate.
-Write the regression first, observe it fail in that clean container, then make
-it pass there. Host-only results are supplemental.
+For Auto Research changes, `npm run test:clean` is the iterative authoritative
+TDD gate. It may reuse input-valid Docker build layers, but every invocation
+runs the tests in a separately created offline container with isolated HOME and
+temporary filesystems. Write the regression first, observe it fail there, then
+make it pass in another fresh container. Host-only results are supplemental.
+
+Run `npm run test:clean:cold` after changing `.dockerignore`, the clean-test
+Dockerfile, a dependency manifest or lockfile, and before delivery. Hosted PR
+and publish workflows use this cold mode explicitly; it adds `--no-cache` but
+does not use `--pull`, because base versions change only through reviewed digest
+updates.
 
 Use `npm run typecheck` for a faster TypeScript-only check.
 Use `npm run prepush:gate` when `docpact` is installed and you want the

package/README.md

@@ -12,8 +12,8 @@ checkPaths:
   - package.json
   - bin/**
   - src/**
-lastReviewedAt: 2026-08-12
-lastReviewedCommit: ac34420a63cb02898f8c38b1c1af64f0439e862f
+lastReviewedAt: 2026-08-16
+lastReviewedCommit: b8c2d6acaf44de0b9c7ea4530e54baff84ad034b
 ---
 
 # Tiangong AI CLI
@@ -269,7 +269,7 @@ regular non-symlink `runtime-lock.json` exact stable CLI version. Setup and
 release CI reject a missing resolver or any stale exact CLI version in the
 orchestrator's `SKILL.md` or `references/*.md`.
 
-### Top-journal Research Policy and final publication gate
+### Top-journal Policy, scientific design, and publication gates
 
 A `top-journal` project starts with a human-reviewed Markdown Policy, not with
 model execution. After project-scoped setup reaches `READY`, use the guided
@@ -292,22 +292,86 @@ binds the manifest and every document by SHA-256; edits, manifest tampering, or
 expiry block preflight and all later stages until the Policy is reviewed and
 approved again.
 
-Use the Policy project ID when admitting the research project:
+Before search, the current native Codex or Claude host must author a
+project-specific scientific design. The CLI owns the closed schema and rejects
+designs that confuse model-to-model disagreement with observed truth, inflate
+independent sample size through resampling, omit quantity/threshold semantics,
+leave blocking gaps unresolved, or cannot fit the complete review lifecycle.
+The CLI validates, freezes, hashes, and routes this design; it does not author
+the design or launch a nested producer.
+
+Hash binding alone does not make a model executable. Each model declares raw
+implementation bytes, a retrievable safe locator and entrypoint, exact
+environment-lock bytes, implementation/environment status, and a freeze gate.
+Source-derived uncertainty states also declare whether their values are frozen
+or pending, and every joint state maps exact parameter-state IDs. Pending model,
+environment, or uncertainty objects are allowed only when a planned Policy rule
+owns the same due gate. They are exposed in every earlier review packet as
+`futureGateObligations` and become blocking mechanical errors at that gate.
+Freezing them requires a new authoritative generation; it never upgrades the
+old object in place.
+
+Use the same Policy project ID and exact design when preflighting and admitting
+the research project:
 
 ```bash
+tiangong-ai research schema show scientific-design --json
 tiangong-ai research project preflight \
   --question "A specific, testable research question" \
   --goal top-journal --policy-project top-journal-paper \
   --requirements /absolute/path/to/evidence-requirements.json \
+  --design /absolute/path/to/scientific-design.json \
+  --workspace /absolute/path/to/workspace --json
+tiangong-ai research project init top-journal-paper \
+  --question "A specific, testable research question" \
+  --goal top-journal \
+  --requirements /absolute/path/to/evidence-requirements.json \
+  --design /absolute/path/to/scientific-design.json \
+  --design-producer-agent codex \
+  --design-producer-session OPAQUE_NATIVE_SESSION \
+  --confirm-budget \
   --workspace /absolute/path/to/workspace --json
 ```
 
 The base evidence lifecycle remains
 `discover -> acquire -> analyze -> synthesize -> review -> close`, authored in
-the current interactive Codex or Claude Code host. After base closure, the
-current native host writes a final manuscript and schema-valid publication
-assessment. `research publication freeze` then content-addresses the Policy,
-evidence snapshot, base outputs, manuscript, assessment, and supplements.
+the current interactive Codex or Claude Code host. A fresh independent reviewer
+must first pass three hash-bound scientific gates: `research-design` before
+discovery, a real-record and outcome-blind `evidence-construct` canary after
+discovery and before acquisition, and `pilot-methods` after acquisition and
+before analysis. Reviewer prose cannot override a mechanical failure.
+
+```bash
+tiangong-ai research schema show scientific-assessment-research-design --json
+tiangong-ai research project scientific review prepare top-journal-paper \
+  --role research-design \
+  --assessment /absolute/path/to/research-design-assessment.json \
+  --reviewer-agent claude \
+  --reviewer-session FRESH_OPAQUE_REVIEW_SESSION \
+  --workspace /absolute/path/to/workspace --json
+tiangong-ai research schema show scientific-review-research-design --json
+tiangong-ai research project scientific review submit top-journal-paper \
+  --role research-design --review /absolute/path/to/review.json \
+  --workspace /absolute/path/to/workspace --json
+```
+
+Repeat the same prepare/submit route for `evidence-construct` and
+`pilot-methods` at their stage boundaries. A top-journal fork or addendum is a
+new authoritative generation and therefore requires a target-specific approved
+Policy, design, and fresh native producer session; it cannot inherit scientific
+approval from a superseded generation.
+
+Review packet `stageInputs` identify promoted portable objects by purpose,
+owner, source locator, and SHA-256 over raw file bytes. `packetSha256` is the
+logical packet identity that excludes its own identity field; the portable
+audit manifest separately records the raw stored packet-file digest. This keeps
+packet identity and byte-level transfer verification explicit rather than
+overloading one hash with both meanings.
+
+After base closure, the current native host writes a final manuscript and
+schema-valid publication assessment. `research publication freeze` then
+content-addresses the Policy, scientific design and early reviews, evidence
+snapshot, base outputs, manuscript, assessment, and supplements.
 Exactly four fresh independent sessions review that frozen generation:
 evidence, methods/reproducibility, domain/novelty, and journal-editor. A revised
 manuscript invalidates prior reviews; reviewer-session reuse is rejected from
@@ -332,6 +396,21 @@ The CLI returns a mechanically bounded ceiling:
 `target-journal-submission-ready`. Evidence and review failures can only lower
 it. None of these states predicts or guarantees editorial acceptance.
 
+Before external handoff or archival, export and independently verify a portable
+audit directory. It contains the selected project, portable copies of admitted
+inputs, formal evidence and artifact bytes, Policy/design/review objects,
+outputs, environment fingerprints, and journal proofs. Credentials, setup
+sources, browser profiles, native active state, capsules, unrelated projects,
+and host-specific absolute paths are excluded.
+
+```bash
+tiangong-ai research project audit export top-journal-paper \
+  --output /absolute/path/to/new-audit-directory \
+  --workspace /absolute/path/to/workspace --json
+tiangong-ai research project audit verify \
+  --bundle /absolute/path/to/new-audit-directory --json
+```
+
 `research setup status --json` reports credential persistence separately from
 readiness. It also reports the effective exact-npx CLI package/version/root,
 the selected project orchestrator, any temporary recovery shim, ignored global
@@ -469,6 +548,42 @@ names its registered parent and inherits that parent's canonical source URL;
 it does not invent a second network-download binding, and a conflicting URL is
 rejected.
 
+For top-journal work, the frozen scientific design maps every required evidence
+role to all applicable lawful acquisition routes in the configured environment.
+Every declared agent route for a required role is mandatory, and every required
+capability must map to an available locked broker route at preflight.
+Each broker call carries its exact `acquisition_route_id`; native activity and
+download records carry `acquisitionRouteId`. A missing or mismatched route ID is
+rejected rather than becoming evidence that a method was tried.
+
+Inspect the live, hash-verified route state before declaring a material evidence
+ceiling:
+
+```bash
+tiangong-ai research project access status gpu-resource-impact \
+  --workspace /absolute/path/to/workspace --json
+```
+
+Once all agent routes are terminal, the command first recommends assessing
+required evidence-role coverage. Its `ifEvidenceStillInsufficient` field is a
+conditional access/scope action, not a claim that purchase is always necessary.
+
+Successful broker/native/download completion, explicit broker authentication or
+entitlement denial, and validated deterministic no-OA download outcomes can be
+terminal. HTTP 422, malformed requests, configuration errors, timeouts, 429,
+5xx, cancelled downloads, and login/MFA/CAPTCHA/security challenges are not
+route exhaustion. Challenges pause immediately through an
+`interactive-challenge` handoff.
+
+Only after every required plan-bound agent route for a still-missing required
+evidence role has exact terminal event hashes may the native host submit a
+schema-v2 `evidence-exhausted` handoff. The durable handoff names each remaining
+purchase, subscription, institutional authorization, owner input, external data
+request, or field collection action with an official non-sensitive locator and
+resume criteria. Research then stops; it does not spend more budget on
+low-yield substitutes. If no lawful remaining route exists, the user must narrow
+or abandon the unsupported scope before a new reviewed generation can resume.
+
 Successful acquisition freezes an immutable evidence snapshot before analysis.
 The reviewer and mechanical closure bind and recheck the snapshot chain,
 ledger, receipts, selected artifacts, excerpts, analysis, and report. Refresh a
@@ -527,12 +642,15 @@ repair with no research tools.
 Total, per-package, output, repair, broker-response bytes, estimated broker
 context tokens, context items, wall-time, output-count, output-size, and attempt
 limits live in `.tiangong-research/config.json`.
-New production workspaces use generous but finite runaway ceilings: 20,000,000
-total tokens and package ceilings of 12,000,000 for discovery, 2,000,000 for
+New production workspaces use generous but finite runaway ceilings: 50,000,000
+total tokens, USD 5,000, 30 days, and package ceilings of 12,000,000 for discovery, 2,000,000 for
 acquisition, 1,500,000 each for analysis and synthesis, and 2,500,000 for
 review. Primary output is bounded at 32,000 tokens and a separately invoked
 repair at 16,000. The production broker hard ceiling is 256 bounded views with
 32,000 context tokens per view; input context is bounded at 128,000 tokens.
+Top-journal admission additionally reserves three early scientific reviews at
+500,000 tokens each, four final publication reviews at 750,000 each, and one
+4,000,000-token revision cycle, including their finite wall-time allowances.
 These values are not a target spend. Coverage-derived working plans and early
 stop control ordinary use, while the finite ceilings, three attempts per
 package, and explicit confirmation above the cost threshold stop runaway work.