docs(docs): R5.6 M1 corrected, final rehearsal results, audits, memory

This commit is contained in:
bchanot committed 2026-10-25 02:10:38 +02:00
1 parent baff88280f
commit 59521dc195
5 files changed
+191

No files matched your search

@@ -0,0 +1,127 @@
# Security audit: R5.6 final switch state (M1 correction)
- Date: 2026-10-25. Mode: audit, fresh and blind (no prior verdict used).
- Target: worktree `/tmp/claude-1000/-home-bchanot-Documents-faunosteo/3571c861-cd6e-4751-a25e-c26da95beb85/scratchpad/r56final`,
detached at 8e77e37 + `.claude/tasks/plans/r5-6-switch.patch`. The worktree diff is line-identical to the committed
patch at baff882 (index lines aside). No code under `crates/ packages/ apps/ scripts/ supabase/` differs between
943a25a, 8e77e37 and baff882, so the patch applies to the official code state.
- Contract: `.claude/tasks/contracts/2026-10-25-rust-r5-6-rehearsal.md` S1-S8 and the 2026-10-25 Clarification.
- Scope: the 30 files of the diff (9 Rust sources, Cargo.toml, 7 Rust tests + 2 fixtures, 4 TS tests/helpers,
2 scripts, package.json, .prettierignore, CLAUDE.md, the 2 switched seed files).
## Attack on the alias decision
Every Rust site that decides "same document" was enumerated (grep of `equivalence`, `TRUSTED_ALIASES`,
`TrustedAlias`, `same_document`, `recognises_historical_stamp`, `document_sha256` across `crates/core/src`,
`crates/wasm/src`, `apps/app/src`, `packages/core-wasm/src`).
| Path | Decision | Result |
|---|---|---|
| `ops.rs:361` `archive.snapshot` → `snapshot_rules.rs:60-68` `build_sheet_snapshot` | alias only through `recognises_historical_stamp` | holds |
| `pdf/write/input.rs:204-210` `assert_exportable` (wasm export session, `export_pdf`, `export_from_envelope`) | strict `sheet_id` + `document_sha256` equality, unchanged from 943a25a | holds |
| `report/rebase.rs:42-46` (reached by `draft.restore`, `archive.import`, `import_pdf`, `restore_copy`) | exact equality (pre-existing) or `recognises_historical_stamp` with same sheet id | holds |
| `archive/read.rs:116-119` `snapshot_matches` | strict equality report vs archived snapshot, unchanged | holds (mixed alias/canonical archive loses its snapshot, tested) |
| `pdf/compare/lines.rs:350-376` report-compare | `recognises_historical_stamp`, both raw values printed through `esc` | holds |
| `session.rs`, `ops.rs`, `facade.rs` (one arm), `crates/wasm` | no own comparison; route to the above | holds |
| app TypeScript | no stamp comparison (only `provider.ts:52` copies the ref) | holds |
`recognises_historical_stamp` (`loaded.rs:163-171`) requires, in this order: a loader-verified `v1` value, that value
still equal to `reference.document_sha256`, the registry pair (sheet id, stamp as alias, verified value as canonical,
one direction only, `fingerprint.rs:63-75`), and a fresh `document_fingerprint(&self.document)` equal to the verified
value. Because the registry canonical must equal a value recomputed from the current document, the privilege is
granted only when the loaded content hashes to the registered equine canonical (SHA-256 preimage otherwise). A
JS caller building a `SheetHandle` from a hand-made bundle gains nothing unless its document is that content.
Forging the private state:
- Struct literal and functional update `..base`: impossible outside `loaded.rs` (private `verified_fingerprint`,
`index`, `drawings`; E0451). No other constructor than `from_bundle` (grep: test helpers all go through it).
- `Deserialize`, `Default`, `Serialize`: not derived on `LoadedSheet`. `unsafe_code = "forbid"` workspace-wide
(`Cargo.toml:55`), no transmute. No `Cell`/`RefCell`.
- `Clone`: copies a consistent state; mutating the clone's pub `document` or `reference` is caught by the
recomputation and the `verified == reference` check (tested, `r5_6_switch.rs` `a_mutated_document_or_reference_never_keeps_the_alias`).
- `SheetHandle` (`crates/wasm/src/lib.rs:26-37`) holds the sheet privately; no `&mut` access to it from JS.
- Even a forged `verified_fingerprint` on a v0 sheet would only grant the alias when the content is the canonical
one: a policy deviation from S3, never a content confusion.
Fingerprint block handling (`loaded.rs:39-57`, `191-210`): absent → v0, unverified, no privilege; explicit `null` →
refused (`present_block`); string, unknown key, missing key → refused (`deny_unknown_fields`, required fields);
convention other than `v1`, non 64-lowercase-hex value, value equal to a registered alias, unregistered alias in the
list, value not recomputed from the document (including a document with integrity errors) → refused at load,
closed `invalid_sheet` kind, detail never rendered (`useSheet.ts:60`). Duplicate JSON key: last wins, a trailing
`null` is refused. A misspelt block key at `ref` level (no `deny_unknown_fields` on `BundledRef`) loads as v0:
fail-closed (less privilege, old reports rebase onto identical content).
v1 with warnings (`fingerprint.rs:253-256`): errors refuse, warnings do not. This only widens which documents
get a value; it grants no equivalence, and a warning-only document with another content gets another value
(tested, it rebases old reports).
## Recursion and cost
- `canonical_json` (`fingerprint.rs:121-193`): containers refused at depth ≥ 128 with `CanonicalError::TooDeep`,
tested at the bound (`sheet_fingerprint.rs:280-292`). Its only non-test caller is `document_fingerprint`, whose
only caller is `loaded.rs` on the loaded sheet's document (the build-time bundle). Attacker data (archives,
drafts, PDFs, rescue copies) never reaches `canonical_json`; archived snapshot finding depth stays capped at 8
by `assert_snapshot` before shape. The new load-time `validate_sheet_document` + `serde_json::to_value` on the
bundle are bounded by serde_json's parse limit (127 containers).
- Recomputation cost: native release, a test with about 3 fingerprints, 7 loads and 1 import runs in 84-87 ms
against 32 ms for 3 loads with none, so one fingerprint of the 296 KB document is in the low milliseconds.
The recomputation runs only when the stamp already equals the registered alias (short-circuit order,
`loaded.rs:167-170`; identical stamps never pay it), at most once per `archive.snapshot` (once per export,
`export.ts:146`), per import (one candidate, `import.rs:24-30`) or per draft restore, plus one at each v1
load. Import worker timeout 20 s (`controller.ts:19`). Repeated calls only cost the user's own tab: no DoS.
## Findings
No CRITICAL, no HIGH, no new security suppression.
- [LOW] hardening — `crates/core/src/domain/sheet/fingerprint.rs:63` — `equivalence` is a crate-public function
that answers `TrustedAlias` from the registry alone, with no verified state and no recomputation. Its only
caller today is `recognises_historical_stamp` (`loaded.rs:169`); a future caller (the Supabase sheet provider
of Step 13, a new export path) could reopen the flaw of the previous round. Hint: `pub(crate)` or private to
`loaded`, or a census test that pins its callers.
- [LOW] by design, context — `crates/core/src/pdf/write/input.rs:204-210` — the export session accepts any
self-consistent caller-supplied (report, snapshot) pair: a script in the page can write an archive whose report
and snapshot both carry the alias with a hand-built snapshot. No equivalence is granted there (every reader
re-decides against its loaded sheet), the alias adds nothing a forged canonical stamp could not already do,
and this is the pre-switch behaviour for any stamp.
- [LOW] context, pre-existing — `crates/core/src/domain/report/rebase.rs:43` — the exact-equality branch keeps a
report of another sheet id unrebased when it carries the loaded value; unchanged since R2 (the new alias branch
does check the sheet id). Equal SHA-256 means equal content, so no confusion.
- [LOW] context — `crates/core/tests/r5_6_switch.rs:8` — new `#![allow(clippy::disallowed_methods,
clippy::unwrap_used, clippy::expect_used)]`, test-only, the header 40 existing test files carry; not a
security-tool suppression.
- Out of diff, filtered: guardrails `WARN SB-POLICY-NOT-OWNER`
`supabase/migrations/20261008110000_anatomical_sheets.sql:1177` (`species_select_active`, catalog read
policy), pre-existing.
Non-security note: `r5_6_switch.rs:133-135` writes `<repo>/target/tmp/r5-6/` even when `CARGO_TARGET_DIR` points
elsewhere (git-ignored, `.gitignore:65`).
## Proof
- semgrep 1.168.0, `p/security-audit`, `p/secrets`, `p/owasp-top-ten`, `--metrics=off`: 163 rules on the 30 diff
files, 0 results, 0 errors. Caveat: these packs carry few Rust rules; the Rust review above is manual.
`p/*` content is fetched at run time, not frozen by the CLI pin.
- Checklist, 9 checks on the added lines and the tree: hardcoded secret/auth URL (0), SQL by concatenation (0),
unsanitised render/eval (0), sensitive endpoint without authz (n/a, no endpoint in the diff), internal error to
the user (0: `invalid_sheet` detail not rendered), secret/PII in a log (0), tracked `.env` (0, only
`.env.example`), guardrails `--domain api,supabase` (0 FAIL, 1 WARN out of diff, filtered), new suppressions (0
security; 1 test lint allow, LOW).
- Hermetic tests run in the worktree, `CARGO_TARGET_DIR=<worktree>/target`, release, `--features cli`:
`r5_6_switch` + `sheet_fingerprint` 25 passed; `pdf_import`, `pdf_compare`, `archive_rules`, `report_rules`,
`sheet_contract` 78 passed.
- Seed check: `draft/document.json` still hashes (bytes) to the registered alias 27c7…, `document.json`,
`horse.svg`, `equine_draft.sql` unchanged; only `sheet-ref.json` and the map's `sheet.documentSha256` changed.
- Nothing edited in either tree except this report; no network beyond the semgrep rule download, no Supabase.
TOOL: semgrep 1.168.0 — p/security-audit, p/secrets, p/owasp-top-ten
SCOPE: 30 files
BLOCKING:
(none)
REPORTED (non-blocking):
- [LOW] hardening: crate-public registry-only `equivalence` — crates/core/src/domain/sheet/fingerprint.rs:63
- [LOW] by design: export accepts caller-supplied consistent snapshot — crates/core/src/pdf/write/input.rs:204
- [LOW] pre-existing: exact-equality rebase branch without sheet id — crates/core/src/domain/report/rebase.rs:43
- [LOW] test lint allow (crate convention) — crates/core/tests/r5_6_switch.rs:8
PROOF: semgrep 163 rules on 30 files → 0 findings; checklist 9 checks → 4 findings (LOW), 1 out-of-diff WARN filtered
SECURITY — VERDICT: PASS
@@ -0,0 +1,14 @@
# R5.6 final rehearsal — fresh verifier (Opus), persisted by the orchestrator (verifier is read-only)
Contract `.claude/tasks/contracts/2026-10-25-rust-r5-6-rehearsal.md` (S1-S8 + 2026-10-25 Clarification); final worktree r56final (8e77e37 + patch at baff882).
VERIFY — VERDICT: CONFORME (8/8)
- S1 MET: decision in snapshot_rules.rs build_sheet_snapshot via recognises_historical_stamp (private verified value, reference equality, registry one-way, recompute now); 9 scenarios tested (r5_6_switch 14/14, vitest N-1 2/2, app 2/2); rollback: unpatched build's report-compare reads the re-export, snapshot kept
- S2 MET: input.rs, archive/read.rs, session.rs, ops.rs, crates/wasm, n-1 unchanged vs 943a25a; equality strict
- S3 MET: probe through facade::load_sheet: v0 canonical = no privilege; null/string/{}/array/missing/extra/v0/upper/alias/63-hex/other/forged/self-alias/content-changed/foreign-id refused; warnings-only v1 loads verified
- S4 MET: only the two seed files differ; document.json, horse.svg, equine_draft.sql byte-identical; no oracle/golden/ledger/n-1 in patch; original contexts + post-switch twins
- S5 MET (12-hex prefixes, the tool's style); S6 MET; S7 MET (main tree clean, seed 27c7, patch applies); S8 MET (rust:check rc 0, 1475 passed; check:all 01:53:59 cited)
- Direct-API probe: v1 + old report → 27c7/27c7 exported; v0 canonical → SheetMismatch; mutated document / reference value / reference id → refused; other sheet id → refused; forged stamp → refused; canonical_json 128 OK, 129 TooDeep, 5000 TooDeep without overflow
- Observations: stale comments fingerprint.rs (6, 15, 246-247); RUST.md §11.20 superseded mechanism (superseded by §11.21); null block → invalid_shape in facade (refusal either way)
- FLOOR clean
+4
View File
@@ -362,3 +362,7 @@ R5.4: `draft/{labels,changelog,review_report,drawing,overlay,app_review,svg_comp
- M1 in input.rs; equivalence one-way + fingerprint_verified (v1 recomputed at load, sha2 in wasm). Goldens reproduced byte-exact on switched sheet.
- Original contexts via pre-switch fixtures (O7, O10, R2, ledger, TS + app goldens); check:all 00:56:22 green switched.
- Slip: `pnpm format` rewrote append-only audits in worktree → reverted. Never format over .claude/.
## 2026-10-25 (R5.6 M1 fix)
- Cause M1: export sees only snapshot. Fix in build_sheet_snapshot via recognises_historical_stamp (private verified + recompute at use); input.rs strict, unchanged.
- null block refused, warnings allowed (is_publishable), canonical depth 128. Fresh worktree final: all gates green, security PASS, verifier CONFORME 8/8. STOP before activation.
+1
View File
@@ -347,6 +347,7 @@ feat/findings-tree-v2 merged (FindingsTreePicker tracked in HEAD); all work belo
- [x] M1 option 1 (input.rs), directional equivalence + fingerprint_verified, v0/v1 load rules, report-compare status, generator order
- [x] original contexts (O7, O10, R2 Zod, R5.1 ledger, TS golden, app goldens) + separate post-switch tests; rust:check + check:all 00:56:22 green on switched state
- [x] gates: security PASS (1 MEDIUM, 6 LOW), verifier ECARTS(3) = export grants alias without fingerprint_verified (fix needs snapshot_rules.rs, outside scope) → checkpoint, STOP, owner decision
- [x] R5.6 M1 correction + full rehearsal (owner 2026-10-25; snapshot_rules.rs authorised): recognises_historical_stamp, null block refused, warnings allowed, depth 128; fresh worktree r56final: rust:check + check:all 01:53:59 green; security PASS (LOW); verifier CONFORME 8/8 → checkpoint, STOP (activation on signal)
- [ ] R5.6 D-R5 switch (convention, aliases, membership test in gated readers with per-file go, seed regenerated once by the Rust tool, `.prettierignore`, CHANGELOG)
- [ ] R5.7 consumer census, domain fixtures moved, packages deleted, xmldom + prettier out of the lockfile, docs, ledger; `pnpm check:all`; merge only on the owner's signal
- [ ] DEV hardening (owner 2026-10-24, risk accepted temporarily, not scheduled): Supabase on CLI demo keys + 54321 LAN-open (rootless, UFW allow incoming, no 54321 rule) → own JWT secret/signing keys via env() or narrower 54321 rule before any untrusted device or user (ENVIRONMENTS.md §3.6); never these keys in production; `confirm_sheet_asset_upload` object check before public exposure
+45
View File
@@ -3590,3 +3590,48 @@ an alias. Not applied: awaiting the owner. LOW kept: `"fingerprint": null` and a
(no privilege), public mutable `reference`/`fingerprint_verified` fields, a draft-allowed validation warning
would make a `v1` bundle unloadable, `canonical_json` recursion unbounded (trusted input today), stale
comments on `pnpm import:equine` in `scripts/import-equine.sh` and CLAUDE.md.
### 11.21 R5.6 M1 corrected, full rehearsal repeated (2026-10-25)
Owner brief « R5.6 : autorisation de correction M1, nouvelle répétition obligatoire ». Supersedes the M1
mechanism of §11.20: `input.rs` `keep_the_report_stamp` is gone and `pdf/write/input.rs` is unchanged
from 943a25a (strict equality). Final patch `.claude/tasks/plans/r5-6-switch.patch` (30 files, baff882),
rehearsed in a fresh detached worktree of 8e77e37.
**Cause of the defect.** The export only sees the snapshot (`ExportInput`), never the loaded sheet, so
it could not know whether the sheet's `v1` value had been verified; it granted the registered pair on the
value the snapshot claimed (verifier probe: a v0 sheet carrying the canonical value exported an old
report).
**Fix.** The decision is taken where the loaded sheet is known: `domain/archive/snapshot_rules.rs`
`build_sheet_snapshot` (behind `archive.snapshot`) keeps the report's historical stamp only when
`LoadedSheet::recognises_historical_stamp` holds: the loader verified a `v1` block (private
`verified_fingerprint`, set only by `from_bundle`), that value still equals the reference, the registry
binds the stamp to it for this sheet id (one direction), and the current document recomputes to it at
the moment of the decision (so a mutated pub `document` or `reference` loses the privilege). The rebase
and `report-compare` use the same predicate; the export refuses any report whose stamp differs from the
snapshot's. `"fingerprint": null` and malformed blocks are refused (`present_block`); a `v1` fingerprint
refuses integrity errors and accepts draft warnings (`is_publishable`: the two warning rules are
allowed in a draft and never block publication; a warnings-only `v1` sheet loads); `canonical_json` is
bounded at 128 levels (`CanonicalError::TooDeep`, serde_json's parse limit). Public API: the patch's own
pub field `fingerprint_verified` becomes the method `is_fingerprint_verified()`; no released or wasm API
changes.
**Gates on the exact final state.** `pnpm rust:check` green; `pnpm check:all` green 01:53:59; `--check`
0 differences; the Rust `--out` equals the switched seed for `sheet-ref.json` and
`report-migration-map.json` and leaves `document.json`, `horse.svg` and `equine_draft.sql` byte-identical.
`r5_6_switch` 14 (direct export API: unverified v0 refused, mutated document or reference refused,
another sheet id refused, contradicting `v1` content refused at load, `null` and malformed blocks refused,
warnings-only `v1` loads, the earlier scenarios), `sheet_fingerprint` 11 (depth 128/129), vitest
`r5-6.n-1` 2 (snapshot from the core's `archive.snapshot`; the v0 counterpart refused), app goldens and R4
digests byte for byte. Security PASS, LOW only (`.claude/audits/2026-10-25-rust-r5-6-final-security.md`);
fresh verifier CONFORME 8/8 with its own direct-API probe (`.claude/audits/2026-10-25-rust-r5-6-final-verify.md`).
Wasm: 1,532,324 B raw / 518,018 B gzip before, 1,575,574 / 531,715 after (+43 KB raw, +13.7 KB gzip: `sha2`,
the canonicaliser, the load check); no wasm size budget is defined in the repository.
Residual, not changed after the last gate: `equivalence` is public and grants from the registry alone (its
only caller is the proof predicate; make it crate-private at activation); stale comments in
`fingerprint.rs` (lines 6, 15, 246-247: "zero integrity issue", "will declare"); a page script can still
hand the export a self-consistent report and snapshot (no equivalence results, every reader decides again);
the rebase's exact-equality branch has no sheet-id check (since R2); `r5_6_switch` writes its re-export under
`<repo>/target/tmp` regardless of `CARGO_TARGET_DIR`.