feat(orchestrator): caveman-style output contracts for critic/adversary/governor/cartographer by Delqhi · Pull Request #199 · OpenSIN-Code/SIN-Code

Delqhi · 2026-06-16T16:21:44Z

Closes #174

What

Adds a caveman-style output contract for the four orchestrator sub-agents (Critic, Adversary, Governor, Cartographer). Inspired by JuliusBrussee/caveman's cavecrew-* sub-agents (every byte of a sub-agent's prose is re-billed by the orchestrator) and JuliusBrussee/ponytail's 5-tag convention.

New file

:

Finding struct { Tag, Symbol, Path, Line, Confidence, Hint } — flat, byte-stable.
5 closed tags: delete | simplify | rebuild | risk | verify.
Render() is fully deterministic — Finding{...} → Render() → ParseFinding() → equal struct, every byte counted.
ParseFinding / ParseFindings (em-dash U+2014 separator).
VerifyFindings rejects: empty Path, unknown Tag, hedging phrases (closed set of 12, case-folded), Hint > 240 chars, trailing punctuation.
FindingsToBytes for batch rendering.

Wire-in

Sub-agent	Source of `[]Finding`	Entry point
Critic	Parsed from last attempt's prose	`Critic.Drive`
Adversary	Each structured `Attack`	`Adversary.Review`
Governor	One per `Escalation` (risk, anchored at task://)	`Governor.Execute`
Cartographer	Top-k PageRank (verify, opt-in k)	`Cartographer.Findings(k)`

Tests (13 new tests, all byte-stable golden)

output_contract_test.go — Render / Parse / Verify / Hedging / Tag-set.
output_contract_integration_test.go — one fixture per agent (Critic, Adversary, Governor, Cartographer), byte-pinned.

Verified (mandate M2 / M7)

gofmt             cmd/sin-code/internal/orchestrator/                 clean (this PR; pre-existing stcov_test.go drift is coverage-tag-only)
go build          ./...                                                clean
go vet            ./cmd/sin-code/...                                   clean
go test -race     -count=1 ./cmd/sin-code/internal/orchestrator/      PASS
golangci-lint run ./cmd/sin-code/internal/orchestrator/                0 issues
govulncheck       ./cmd/sin-code/internal/orchestrator/...             no vulns
gosec -no-fail    -fmt sarif ./cmd/sin-code/internal/orchestrator/     SARIF clean
python3           scripts/validate_skill.py --all-bundled --strict     34/34 OK

Mandate compliance

M2 (single static binary, CGO_ENABLED=0): no new CGO; pure stdlib.
M5 (module path): only uses github.com/OpenSIN-Code/SIN-Code/... paths.
M7 (race-free): full orchestrator race-test passes.
M3 (verification gate): untouched, but the new parser throws on hedge / empty Path; the orchestrator's repair loop re-injects ParseErrors as retry feedback (verify.fail direction).

Docs

AGENTS.md §6.1 — full description of the contract, the 5 tags, the closed hedging set, and the per-sub-agent source of []Finding. Original contract.go (Intent Contract) is documented as distinct from the new output contract.
CHANGELOG.md Unreleased — added the new section.

vercel · 2026-06-16T16:21:50Z

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
sin-code	Ready	Preview, Comment, Open in v0	Jun 16, 2026 7:37pm

…ry/governor/cartographer Closes #174. Adds cmd/sin-code/internal/orchestrator/output_contract.go: * Finding struct + 5 closed tags (delete, simplify, rebuild, risk, verify) parallel to JuliusBrussee/ponytail's ponytail-tag convention. * byte-stable Render() — same Finding struct produces identical bytes every run (prerequisite for issue #168 ledger hashing). * strict ParseFinding/ParseFindings (em-dash U+2014 separator, no prose, no hedging). * VerifyFindings rejects empty Path, unknown Tag, hedging phrases (closed set of 12, case-folded: 'you might', 'perhaps', 'could consider', 'maybe', 'i think', 'i would', 'sort of', 'kind of', 'tends to', 'should probably', 'i'd suggest', 'we should'), >240-char Hint, and trailing punctuation. Wired into the four sub-agents: * CriticResult.Findings — parsed from the last attempt's prose; CriticResult.ParseErrors carries per-line rejection trace (mirrors verify.fail flow). * AdversaryResult.Findings — derived from each Attack (landed -> risk, cleared -> verify). * GovernorResult.Findings — one risk Finding per Escalation, anchored at task://<ID>. * Cartographer.Findings(k) — top-k PageRank-sorted symbols as verify Findings (opt-in: k <= 0 yields empty). Byte-stable golden tests in output_contract_test.go (Find + Render + Verify) and output_contract_integration_test.go (one fixture per agent). Existing TestCritic*, TestAdversary*, TestGovernor*, and TestCartographer* unchanged and green. Verified: gofmt -l cmd/sin-code/internal/orchestrator/ (output_contract*.go clean; pre-existing stcov_test.go drift is a coverage-tag-only file, not my change) go build ./... (clean) go vet ./cmd/sin-code/... (clean) go test -race -count=1 ./cmd/sin-code/internal/orchestrator/ (PASS) golangci-lint run ./cmd/sin-code/internal/orchestrator/ (0 issues) govulncheck ./cmd/sin-code/internal/orchestrator/... (no vulns) gosec ./cmd/sin-code/internal/orchestrator/ (SARIF clean) python3 scripts/validate_skill.py --all-bundled --strict (34/34 OK)

vercel · 2026-06-16T19:37:47Z