hermes/cto
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Add Case Stage 3 copied repo PRD

Browse Source
This commit is contained in:
Svrnty 2026-05-31 23:09:33 -04:00
parent ccf45bf338
commit 41de06d9c9
7 changed files with 274 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

86
.sot/03-PROTOCOLS/CTO-CASE-STAGE3-COPIED-REPO-ISSUES.md Normal file
View File

@ -0,0 +1,86 @@
---
name: cto-case-stage3-copied-repo-issues
tier: local
status: draft
owner: jp
source: .sot/03-PROTOCOLS/CTO-CASE-STAGE3-COPIED-REPO-PRD.md
created: 2026-06-01
last_reviewed: 2026-06-01
lifecycle_classification: planning
core_promotion_status: not-promoted
description: Child-local issue sequence for Stage 3 Case copied-repo fixture proof.
---
# CTO Case Stage 3 Copied Repo Issues
Local planning SOT only. Not a Core Protocol. Not active Core authority.
## Issue Sequence
### CTO-WORK-033 - Stage 3 Copied Repo PRD
Type: AFK
Status: validated.
Blocked by: CTO-WORK-012
User stories covered: CTO Case Candidate Backend PRD stories 4, 5, 7, 8, 9, 10, 11, 13.
What to build: Define the Stage 3 copied-repo fixture proof before implementation starts.
Acceptance criteria:
- [ ] PRD states Stage 3 allowed mutation scope is `copied local repository fixture only`.
- [ ] PRD requires Stage 2 validation before Stage 3.
- [ ] PRD requires `CTO_HARNESS_ALLOW_CASE=1` and `CTO_HARNESS_CASE_STAGE=3`.
- [ ] PRD requires source repository ownership and local source classification.
- [ ] PRD requires source repository HEAD and status before and after Case execution.
- [ ] PRD requires copied fixture clean start and clean end proof.
- [ ] PRD forbids Target Repository, source repository, Case source, vendor source, Hermes WebUI, and Cortex Core mutation.
- [ ] PRD requires full Harness Evidence Interface artifacts.
- [ ] PRD requires dirty-starting-tree, dirty-ending-tree, timeout, artifact-write-failure, disallowed-file, failed-tests, and missing-required-event failure fixtures.
- [ ] Local CTO validator checks Stage 3 PRD and issue artifact.
Allowed files: CTO child workspace planning docs and local validator only.
Validator: `python3 tools/validate_cto_child.py`
Done evidence: PRD, issue artifact, validator JSON, clean worktree, commit.
### CTO-WORK-034 - Stage 3 Harness Copied Repo Fixture Route
Type: AFK
Status: blocked.
Blocked by: CTO-WORK-033
User stories covered: CTO Case Candidate Backend PRD stories 4, 5, 7, 8, 9, 10, 11, 13.
What to build: In `/home/svrnty/workspaces/hermes/cto/harness`, implement the Stage 3 copied-repo fixture route behind the existing `case` engine seam.
Acceptance criteria:
- [ ] `case` remains disabled by default.
- [ ] `CTO_HARNESS_ALLOW_CASE=1` remains required.
- [ ] `CTO_HARNESS_CASE_STAGE=3` is required before copied-repo Case execution.
- [ ] Missing Stage 3 gate emits blocked evidence and does not run Case.
- [ ] Source repository is copied before Case starts.
- [ ] Source repository HEAD and status before/after proof match.
- [ ] Case mutates only the copied fixture.
- [ ] Copied fixture starts clean and ends clean after harness post-processing.
- [ ] No Target Repository path is inspected or copied.
- [ ] Required artifacts include source non-mutation proof, clean-start proof, clean-end proof, `report.json`, `report.md`, `events.normalized.jsonl`, `trace.jsonl`, `patch.diff`, `test.log`, and backend logs.
- [ ] Failure fixtures fail closed for dirty source start, dirty copied fixture start, dirty copied fixture end, timeout, artifact write failure, disallowed file, failed tests, and missing required event.
- [ ] Fake remains the default validation lane and broad health remains green after focused Stage 3 validation.
Allowed files: Hermes CTO harness engine, copied-repo fixtures, focused Stage 3 validator, harness docs, and tests. WebUI, Core, Case source, vendor source, source repository, Target Repository, and external developer repositories are forbidden.
Validator: `python3 harness/runner/validate-case-stage3.py --harness-root harness --json`, then `harness/evals/health.sh --json`.
Done evidence: Stage 3 pass report, failure fixture reports, source non-mutation proof, clean-start proof, clean-end proof, artifact digests, clean worktree, commit.
## Granularity Check
This is intentionally two slices: one planning route and one executable harness route. It is not over-granular because Stage 3 introduces source repository copy and non-mutation proof, which are distinct from Stage 2 artificial fixture proof.

109
.sot/03-PROTOCOLS/CTO-CASE-STAGE3-COPIED-REPO-PRD.md Normal file
View File

@ -0,0 +1,109 @@
---
name: cto-case-stage3-copied-repo-prd
tier: local
status: draft
owner: jp
source: .sot/03-PROTOCOLS/CTO-CASE-STAGED-PROOF-GATES.md
created: 2026-06-01
last_reviewed: 2026-06-01
lifecycle_classification: planning
core_promotion_status: not-promoted
description: Child-local PRD for Stage 3 Case copied-repo fixture proof.
---
# CTO Case Stage 3 Copied Repo PRD
Local planning SOT only. Not a Core Protocol. Not active Core authority.
## Problem Statement
Stage 2 proves Case can patch a copied artificial fixture through the CTO Harness. That does not prove Case is safe around a real repository shape. Stage 3 must prove the next narrow behavior: copy an owned local source repository into a fixture workspace, run Case only inside the copied fixture, and prove the source repository remains unchanged.
## Solution
Add a Stage 3 copied-repo fixture route for the Hermes CTO harness. The route uses an owned local source repository only as read-only input. The harness copies that source into a runtime fixture, records source non-mutation proof, runs Case against the copied fixture, and accepts only Harness Evidence Interface proof.
Stage 3 keeps all earlier gates. `case` remains disabled by default. `CTO_HARNESS_ALLOW_CASE=1` and `CTO_HARNESS_CASE_STAGE=3` are required before copied-repo Case execution. Missing gates mean blocked, not warning.
Allowed mutation scope is `copied local repository fixture only`. Writable roots are limited to `runtime_workspace_root` and `run_artifact_dir`. Source repository, Target Repository, Case source, vendor source, external developer repositories, Hermes WebUI, and Cortex Core are read-only or forbidden.
## Scope
- Define one copied-repo fixture proof route.
- Require Stage 2 validated evidence before Stage 3 execution.
- Require source repository ownership and local path classification before copy.
- Require clean source repository status before copy.
- Require clean copied fixture status before Case starts.
- Require source repository HEAD and status proof before and after execution.
- Require copied fixture clean ending tree after harness post-processing.
- Preserve full Harness Evidence Interface artifacts.
- Compare report shape, event order, allowed writes, tests, blockers, digests, and freshness against Stage 2 expectations.
- Add dirty-starting-tree, dirty-ending-tree, timeout, and artifact-write-failure failure fixtures.
## Non-Goals
- Do not mutate a Target Repository.
- Do not mutate the source repository.
- Do not use an external developer repository as source.
- Do not push, merge, deploy, close, or open a pull request.
- Do not resolve license or source admission for real-repo work.
- Do not approve Stage 4, Stage 5, Stage 6, default backend status, WebUI Runtime behavior, or Core promotion.
## Acceptance Criteria
- Stage 3 entry requires Stage 2 validated.
- `CTO_HARNESS_ALLOW_CASE=1` remains required.
- `CTO_HARNESS_CASE_STAGE=3` is required.
- Missing Stage 3 gate blocks before Case starts.
- Source repository is an owned local source, not a Target Repository and not external developer source.
- Source repository clean status is recorded before copy.
- Source repository HEAD and status are recorded before and after Case execution.
- Source repository after-proof matches before-proof.
- Copied fixture is created under the run artifact directory.
- Case receives only the copied fixture path, task contract, allowed paths, forbidden actions, verification command, and evidence expectations.
- Runtime writes are limited to `runtime_workspace_root` and `run_artifact_dir`.
- Copied fixture starts clean after copy and baseline commit.
- Copied fixture ends clean after harness post-processing.
- `report.json` records `backend: case`, `case_process_started`, `source_repository_mutated: false`, `runtime_workspace_root`, `run_artifact_dir`, `changed_files`, `blockers`, `artifact_digests`, and freshness proof.
- Required artifacts include `report.json`, `report.md`, `events.normalized.jsonl`, `trace.jsonl`, `patch.diff`, `test.log`, backend logs, source non-mutation proof, clean-start proof, and clean-end proof.
- Failure fixtures cover dirty source start, dirty copied fixture start, dirty copied fixture end, timeout, artifact write failure, disallowed file, failed tests, and missing required event.
- Fake remains the default validation lane.
- No Stage 3 pass may be used as sandbox-repo, owned-repo, default-candidate, or Core promotion evidence beyond its stated scope.
## Validation
- Focused validator: `python3 harness/runner/validate-case-stage3.py --harness-root harness --json`.
- The Stage 3 validator must require Stage 2 validation first.
- The validator must prove missing Stage 3 gate blocks before `case_process_started`.
- The validator must prove source repository non-mutation with before/after HEAD and status evidence.
- The validator must prove copied fixture clean start and clean end.
- The validator must prove no Target Repository path is inspected or copied.
- The validator must run required failure fixtures.
- The validator must emit compact JSON with `ok`, `checked`, `errors`, pass artifact path, failure artifact paths, and source non-mutation proof path.
- Broader Hermes health must run once after focused Stage 3 validation passes.
- CTO child validator must require this PRD and issue artifact before Stage 3 implementation is governed.
## Risks
- Copied-repo proof can be mistaken for real Target Repository approval.
- Source repository mutation could happen through a leaked path if task contracts are weak.
- Dirty tree handling can hide untracked files or generated artifacts.
- Stage 3 can overfit to one toy repo and miss real-repo policy needs.
## Dependencies
- Stage 2 Case artificial fixture is validated.
- Harness Evidence Interface Contract is validated.
- Case Adapter Contract is validated.
- Case Failure Fixture Matrix is validated.
- Real Case Qwen Stage 2 pass evidence exists.
## Success Definition
Stage 3 is successful when Case changes only a copied local repository fixture, emits full Harness Evidence Interface proof, passes verification, fails closed for required copied-repo failure classes, and proves the source repository was not mutated. Stage 3 does not authorize sandbox, owned-repo, default backend, WebUI Runtime, or Core promotion behavior.
## Challenge Findings
- Accepted: source non-mutation proof must include source HEAD and status before and after Case execution.
- Accepted: clean copied-fixture start and clean copied-fixture end are separate proofs.
- Rejected: running Stage 3 directly on a Target Repository. That skips the proof ladder.

7
.sot/03-PROTOCOLS/CTO-CASE-STAGED-PROOF-GATES.md
View File

@ -130,6 +130,8 @@ Validation evidence:
## Stage 3 - Copied Repo Fixture
Status: planned. Execution remains blocked until `CTO-WORK-034` produces Harness Evidence Interface pass evidence.
Entry gates:
- Stage 2 is validated.
@ -146,6 +148,11 @@ Required artifacts:
- source repository non-mutation proof;
- failure fixture results.
Planning evidence:
- Stage 3 PRD: `.sot/03-PROTOCOLS/CTO-CASE-STAGE3-COPIED-REPO-PRD.md`.
- Stage 3 issues: `.sot/03-PROTOCOLS/CTO-CASE-STAGE3-COPIED-REPO-ISSUES.md`.
Validator expectation:
- all changes occur inside copied fixture;

4
CONTEXT.md
View File

@ -24,3 +24,7 @@ _Avoid_: loose evidence bundle, backend logs, success claim
**Target Repository**:
The owned source repository receiving bounded, approved, evidence-producing code changes.
_Avoid_: vendor source, hidden workspace, disposable scratch by default
**Copied Repository Fixture**:
A runtime copy of an owned local source repository used to prove backend behavior without mutating the source repository or a Target Repository.
_Avoid_: Target Repository, live repo, external developer source

2
README.md
View File

@ -42,6 +42,8 @@ This workspace is registered as a child-local planning workspace. Registration d
| |-- CTO-CASE-STAGE1-GATED-ENGINE-ISSUES.md
| |-- CTO-CASE-STAGE2-ARTIFICIAL-FIXTURE-PRD.md
| |-- CTO-CASE-STAGE2-ARTIFICIAL-FIXTURE-ISSUES.md
| |-- CTO-CASE-STAGE3-COPIED-REPO-PRD.md
| |-- CTO-CASE-STAGE3-COPIED-REPO-ISSUES.md
| |-- CTO-CASE-PROVIDER-ADMISSION-PRD.md
| |-- CTO-CASE-PROVIDER-ADMISSION-ISSUES.md
| |-- CTO-CASE-PROVIDER-BUILD-PRD.md

10
WORKBOARD.yaml
View File

@ -160,3 +160,13 @@ items:
status: validated
source: .sot/03-PROTOCOLS/CTO-CASE-AGENT-PROTOCOL-BLOCKER.md
owner: ""
- id: CTO-WORK-033
title: Stage 3 Copied Repo PRD
status: validated
source: .sot/03-PROTOCOLS/CTO-CASE-STAGE3-COPIED-REPO-PRD.md
owner: ""
- id: CTO-WORK-034
title: Stage 3 Harness Copied Repo Fixture Route
status: blocked
source: .sot/03-PROTOCOLS/CTO-CASE-STAGE3-COPIED-REPO-ISSUES.md
owner: jp

56
tools/validate_cto_child.py
View File

@ -28,6 +28,8 @@ REQUIRED_FILES = [
".sot/03-PROTOCOLS/CTO-CASE-STAGE1-GATED-ENGINE-ISSUES.md",
".sot/03-PROTOCOLS/CTO-CASE-STAGE2-ARTIFICIAL-FIXTURE-PRD.md",
".sot/03-PROTOCOLS/CTO-CASE-STAGE2-ARTIFICIAL-FIXTURE-ISSUES.md",
".sot/03-PROTOCOLS/CTO-CASE-STAGE3-COPIED-REPO-PRD.md",
".sot/03-PROTOCOLS/CTO-CASE-STAGE3-COPIED-REPO-ISSUES.md",
".sot/03-PROTOCOLS/CTO-CASE-PROVIDER-ADMISSION-PRD.md",
".sot/03-PROTOCOLS/CTO-CASE-PROVIDER-ADMISSION-ISSUES.md",
".sot/03-PROTOCOLS/CTO-CASE-PROVIDER-BUILD-PRD.md",
@ -268,6 +270,28 @@ REQUIRED_STAGE2_ISSUE_IDS = [
"CTO-WORK-012",
]
REQUIRED_STAGE3_PRD_PHRASES = [
"Local planning SOT only. Not a Core Protocol. Not active Core authority.",
"Stage 3 must prove the next narrow behavior",
"copied local repository fixture only",
"CTO_HARNESS_ALLOW_CASE=1",
"CTO_HARNESS_CASE_STAGE=3",
"Source repository HEAD and status are recorded before and after Case execution.",
"Source repository after-proof matches before-proof.",
"Copied fixture starts clean",
"Copied fixture ends clean",
"source_repository_mutated: false",
"dirty-starting-tree",
"dirty-ending-tree",
"artifact-write-failure",
"Stage 3 does not authorize sandbox, owned-repo, default backend, WebUI Runtime, or Core promotion behavior.",
]
REQUIRED_STAGE3_ISSUE_IDS = [
"CTO-WORK-033",
"CTO-WORK-034",
]
REQUIRED_PROVIDER_ADMISSION_PRD_PHRASES = [
"Local planning SOT only. Not a Core Protocol. Not active Core authority.",
"https://github.com/workos/case.git",
@ -864,6 +888,28 @@ def main() -> int:
if issue_id not in text:
errors.append(f"missing_stage2_issue_id:{issue_id}")
stage3_prd = ROOT / ".sot/03-PROTOCOLS/CTO-CASE-STAGE3-COPIED-REPO-PRD.md"
if stage3_prd.is_file():
text = stage3_prd.read_text(encoding="utf-8")
if "core_promotion_status: not-promoted" not in text:
errors.append("stage3_prd_missing_not_promoted_frontmatter")
for phrase in REQUIRED_STAGE3_PRD_PHRASES:
checked.append(f"stage3_prd_phrase:{phrase}")
if phrase not in text:
errors.append(f"missing_stage3_prd_phrase:{phrase}")
stage3_issues = ROOT / ".sot/03-PROTOCOLS/CTO-CASE-STAGE3-COPIED-REPO-ISSUES.md"
if stage3_issues.is_file():
text = stage3_issues.read_text(encoding="utf-8")
if "core_promotion_status: not-promoted" not in text:
errors.append("stage3_issues_missing_not_promoted_frontmatter")
if "Local planning SOT only. Not a Core Protocol. Not active Core authority." not in text:
errors.append("stage3_issues_missing_local_planning_notice")
for issue_id in REQUIRED_STAGE3_ISSUE_IDS:
checked.append(f"stage3_issue_id:{issue_id}")
if issue_id not in text:
errors.append(f"missing_stage3_issue_id:{issue_id}")
provider_admission_prd = ROOT / ".sot/03-PROTOCOLS/CTO-CASE-PROVIDER-ADMISSION-PRD.md"
if provider_admission_prd.is_file():
text = provider_admission_prd.read_text(encoding="utf-8")
@ -1059,6 +1105,10 @@ def main() -> int:
checked.append(f"workboard_id:{issue_id}")
if issue_id not in text:
errors.append(f"missing_workboard_id:{issue_id}")
for issue_id in REQUIRED_STAGE3_ISSUE_IDS:
checked.append(f"workboard_id:{issue_id}")
if issue_id not in text:
errors.append(f"missing_workboard_id:{issue_id}")
for issue_id in REQUIRED_PROVIDER_ADMISSION_ISSUE_IDS:
checked.append(f"workboard_id:{issue_id}")
if issue_id not in text:
@ -1108,6 +1158,8 @@ def main() -> int:
"CTO-WORK-027": "validated",
"CTO-WORK-029": "validated",
"CTO-WORK-030": "validated",
"CTO-WORK-033": "validated",
"CTO-WORK-034": "blocked",
}
for issue_id, expected in expected_statuses.items():
checked.append(f"workboard_status:{issue_id}:{expected}")
@ -1134,6 +1186,10 @@ def main() -> int:
errors.append("workboard_missing_stage2_prd_source")
if "CTO-CASE-STAGE2-ARTIFICIAL-FIXTURE-ISSUES.md" not in text:
errors.append("workboard_missing_stage2_issues_source")
if "CTO-CASE-STAGE3-COPIED-REPO-PRD.md" not in text:
errors.append("workboard_missing_stage3_prd_source")
if "CTO-CASE-STAGE3-COPIED-REPO-ISSUES.md" not in text:
errors.append("workboard_missing_stage3_issues_source")
if "CTO-CASE-PROVIDER-ADMISSION-PRD.md" not in text:
errors.append("workboard_missing_provider_admission_prd_source")
if "CTO-CASE-PROVIDER-ADMISSION-ISSUES.md" not in text: