Phase: 05.1 — Resolve Pending TODOs and Integrate deal-stdlib Functionality Audited: 2026年06月09日 ASVS Level: 1 Auditor: gsd-security-auditor (claude-sonnet-4-6) Result: SECURED — 11/11 threats closed

Accepted. Rationale is reasonable:

• Commit 814b63e (chore(spec): bump gitlink for Wave-A grammar reconciliation) records the full rationale in its body: spec advances from dd3bb78 to 74dfea5, enumerates each change (D-05/I5, D-03/I2, D-01/I1, I6, SD-15a/SD-15b, NC-1), and explicitly notes D-07 exclusion of 7ce59bc.
• Single-maintainer repository; full audit trail in git log.
• No multi-party authorization is required or expected at ASVS Level 1.

Accepted. Rationale is reasonable:

• E25xx CLI test fixtures (mixed_mass_units.deal, unknown_unit.deal, mass_for_voltage.deal) are dimension-wrong but syntactically well-formed; they exercise the sema/dimensional layer, not the parser recovery path.
• Parser robustness against adversarial/malformed input is covered separately by the 50-file malformed corpus (Plan 03 recovery_corpus.zig regression gate, green at 55 files after Phase 05.1).
• Risk is bounded: fixtures are created inline in test functions within controlled tempdir environments; no external fixture file ingestion.

None. All four SUMMARY.md ## Threat Flags sections explicitly state "None — no new network endpoints, auth paths, or schema changes introduced."

1. T-05.1-C01 scope clarification: The threat register cites parser_dealx.zig as the primary file, but the new diagnostic arms in this phase live in parser_deal.zig (annotation body loop at :1916-1939 and verification key loop at :1976-1986). Both files satisfy the invariant. The parser_dealx.zig loop at line 834 also advances before continue, confirming the pattern holds project-wide.

2. T-05.1-SC tempfile note: tempfile = "3" appears in cli/Cargo.toml but was introduced in a prior phase (git log for Cargo.toml shows it predates Phase 05.1). Phase 05.1 added no new Cargo, Zig, or Python package dependencies.

3. __array__ residual comments: Two occurrences of __array__ remain in source as comments only (src/fmt.zig:766 doc comment, src/json.zig:448 inline comment). These are not executable references; the synthetic encoding is fully removed from all code paths.

Phase: 05.2 — Implement calc/constraint grammar (SD-21/22/23) in Zig compiler + Rust codegen + editor tooling Audited: 2026年06月11日 ASVS Level: 1 Auditor: gsd-security-auditor (claude-sonnet-4-6) Result: SECURED — 13/13 threats closed

Accepted. No new package installs were introduced in Phase 05.2:

• git log -- Cargo.toml Cargo.lock shows no Cargo manifest changes during the phase 05.2 commit range (11852e0..c6e990e).
• tree-sitter-cli version pinned at 0.25.10 — unchanged from prior phases (confirmed: no package.json changes in tree-sitter-deal during this phase).
• Phase 05.2 RESEARCH.md Package Legitimacy Audit explicitly states "none" for all wave plans.

None. Phase 05.2 SUMMARY.md files for Plans 01-06 all report no new network endpoints, auth paths, file access patterns, or schema changes beyond the declared scope of each wave. Plan 04 SUMMARY explicitly states "Threat Flags: None."

1. T-05.2-W2-01 sync-set discrepancy: The threat register states .kw_calc in STATEMENT_SYNC as one of three mitigation elements. The actual STATEMENT_SYNC set (src/parser_deal.zig:409-417) contains only punctuation tokens (semicolon, comma, r_brace, r_paren, r_bracket). kw_calc is not there. It is in canStartDefinition and isKeyword helpers (lines 2643, 2660). The effective DoS protection is provided by the other two declared elements — the three-tier forward progress guard in the else arms of parseCalcBody (lines 1431-1440) and parseConstraintBody (lines 1526-1536) — which prevent infinite loops regardless of what token is encountered. The arena allocator (line 29-30) bounds memory. This is a documentation inaccuracy in the threat description, not a missing mitigation. Threat is CLOSED on the strength of the two elements that ARE present.

2. T-05.2-W2-01 DEFINITION_SYNC gap: kw_calc is also absent from DEFINITION_SYNC (lines 421-442). This means definition-level error recovery (syncToDefinition) will skip over a calc def keyword rather than stopping at it. The top-level parse loop uses canStartDefinition (which includes kw_calc) before deciding whether to sync, so the gap is benign for the top-level loop. No DoS vector identified.

3. T-05.2-W3-01 implementation detail: checkConstraintRefCycle uses the constraint_chain_map (not the specialization graph) for cycle detection. The map is seeded at sema pass 1 (line 510-521 of sema.zig); cycles in the map are detected at pass 2. This mirrors checkSpecializationCycle exactly as documented in the threat mitigation.

4. Human verification item (T-05.2-W5-02 partial): The TextMate grammar contextual-sig pattern was verified via node -e "JSON.parse(...)" and npm run compile/lint/test:unit (9/9 tests pass). Visual rendering in VS Code was approved by the user during Plan 06 Task 3 checkpoint (commit 6c359aa in vscode-deal). This is recorded in 05.2-VERIFICATION.md under human_verification_status: approved_in_session.