findings: F41-F43 Cobrust corroboration (source-surface-leak + device-name-redact + SPOF-build)

CONTRIBUTING.md

@@ -7,7 +7,7 @@ findings, atomic commits, doc-coverage on the same change.
 
 | Type | Where it lands | Bar |
 |---|---|---|
-| New failure mode (F22+) | `reference/failure-modes-catalogue.md` | At least one concrete empirical instance with citation |
+| New failure mode (F31+) | `reference/failure-modes-catalogue.md` | At least one concrete empirical instance with citation |
 | Case study extension | `case-study/<your-project>-multi-agent-experience.md` | Real project, real outcomes, dated entries |
 | Template improvement | `templates/<name>.md` | Backwards-compatible OR new template |
 | Section refinement in SKILL.md | `SKILL.md` | Cite at least one project where this refinement was tested |
@@ -16,7 +16,7 @@ findings, atomic commits, doc-coverage on the same change.
 ## What we will not merge
 
 - Speculative methodology rules without an empirical instance
-- F-pattern proposals that are restatements of existing F1-F21
+- F-pattern proposals that are restatements of existing F1-F30
 - "Tone" rewrites that lose specific examples or evidence
 - Removal of attribution (e.g. dropping "discovered_by" frontmatter)
 
@@ -28,7 +28,7 @@ findings, atomic commits, doc-coverage on the same change.
    doc + code (if any) + cross-references in the same commit.
 3. **For F-pattern additions**: include `## FN — Title`, `**Signal**`,
    `**Root cause**`, `**Evidence**` (cite project + commit SHA or
-   equivalent), `**Rule of thumb**`. Mirror the existing F1-F21 entry
+   equivalent), `**Rule of thumb**`. Mirror the existing F1-F30 entry
    shape.
 4. **For case studies**: use day-by-day or week-by-week structure. Mark
    counterfactuals (`What would have failed without this discipline:`).

README.md

@@ -1,6 +1,6 @@
 # Agent-Driven Software Development (ADSD)
 
-> Methodology distilled from running a 9-week multi-agent Rust compiler
+> Methodology distilled from running a 12-day multi-agent Rust compiler
 > project where AI agents wrote ≥ 70% of the code under human strategic
 > direction.
 
@@ -10,9 +10,9 @@
 ## What this is
 
 ADSD is **not a framework**. It's a documented working style that survived
-contact with reality: ~178 commits, ~2,611 tests, 43 ADRs, 19 findings, 21
-documented failure modes, 2 P0 codegen bugs caught via organic stress test,
-and a 0.1.1 release shipped publicly.
+contact with reality: ~278 commits, ~2,611 tests, 49 ADRs (0001..0048 + 0047a),
+27 findings, 24 documented failure modes, 2 P0 codegen bugs caught via organic
+stress test, and v0.1.2 stable shipped publicly + α Phase F.2 in flight.
 
 ADSD codifies the discipline that kept the multi-agent project coherent:
 ADRs as decision capture, findings as negative-result memory, bilingual
@@ -46,7 +46,7 @@ simple workflows.
 
 After install, invoke via `/agent-driven-development` or let Claude pick it
 automatically based on context — the description-triggered activation fires
-for multi-agent dispatch planning, ADR drafting, F1–F18 failure-mode triage,
+for multi-agent dispatch planning, ADR drafting, F1–F30 failure-mode triage,
 pre-release audit team design, and similar prompts.
 
 ### As a personal skill (fallback, no plugin system)
@@ -74,14 +74,14 @@ agent-driven-development/
 ├── .claude-plugin/
 │   └── marketplace.json                       # Self-hosted single-plugin marketplace catalog
 ├── plugins/
-│   └── agent-driven-development/              # Plugin root (matches marketplace.json source)
+│   └── adsd/                                  # Plugin root (matches marketplace.json source)
 │       ├── .claude-plugin/
 │       │   └── plugin.json                    # Plugin manifest
 │       └── skills/
 │           └── agent-driven-development/      # Skill — auto-discovered by Claude Code
 │               ├── SKILL.md                   # Main methodology document (~36 KB)
 │               ├── reference/
-│               │   └── failure-modes-catalogue.md  # F1-F21 anti-patterns with empirical evidence
+│               │   └── failure-modes-catalogue.md  # F1-F30 anti-patterns with empirical evidence
 │               ├── case-study/
 │               │   └── cobrust-multi-agent-experience.md  # The founding case study (N=1)
 │               └── templates/
@@ -100,16 +100,41 @@ agent-driven-development/
 ## Quick start (after install)
 
 1. Read [`SKILL.md`](./plugins/adsd/skills/agent-driven-development/SKILL.md) for the full methodology (~36 KB, 30 min).
-2. Read [`reference/failure-modes-catalogue.md`](./plugins/adsd/skills/agent-driven-development/reference/failure-modes-catalogue.md) for the F1–F18 anti-patterns you'll likely hit. Don't re-derive them.
+2. Read [`reference/failure-modes-catalogue.md`](./plugins/adsd/skills/agent-driven-development/reference/failure-modes-catalogue.md) for the F1–F30 anti-patterns you'll likely hit. Don't re-derive them.
 3. Read [`case-study/cobrust-multi-agent-experience.md`](./plugins/adsd/skills/agent-driven-development/case-study/cobrust-multi-agent-experience.md) to see ADSD applied in practice (warts and all).
 4. Copy the relevant template from [`templates/`](./plugins/adsd/skills/agent-driven-development/templates/) into your project's `docs/agent/` tree.
 5. Start writing ADRs as decisions actually happen — not speculatively.
 
+## Documentation
+
+User-facing docs are in [`docs/human/`](./docs/human/) (zh + en parallel per ADSD §3 bilingual mandate). Agent-facing meta-conventions for this repo are in [`docs/agent/`](./docs/agent/).
+
+### Bilingual user docs
+
+| Topic | 中文 | English |
+|---|---|---|
+| Getting started — 30-min onboarding | [`docs/human/zh/getting-started.md`](./docs/human/zh/getting-started.md) | [`docs/human/en/getting-started.md`](./docs/human/en/getting-started.md) |
+| Concept map — mermaid diagrams + narrative | [`docs/human/zh/concept-map.md`](./docs/human/zh/concept-map.md) | [`docs/human/en/concept-map.md`](./docs/human/en/concept-map.md) |
+
+### Agent-facing meta-conventions
+
+- [`docs/agent/conventions.md`](./docs/agent/conventions.md) — repo structure, frontmatter contracts, bilingual mandate enforcement, commit message format, identity hygiene (F21)
+
+### Doc coverage gate
+
+`scripts/doc-coverage.sh` enforces ADSD §3 bilingual mandate on this repo itself: every `docs/human/zh/*.md` MUST have a parallel `docs/human/en/*.md`. Run locally before commits:
+
+```sh
+bash scripts/doc-coverage.sh
+```
+
+The script also verifies reference files have YAML frontmatter and ADR files are zero-padded monotonic. This closes ADSD §3 mandate as F20 systemic prevention applied to ADSD itself.
+
 ## Origin
 
 ADSD was extracted from the [Cobrust](https://github.com/Cobrust-lang/cobrust)
 project, a Rust-implemented Python successor with an AI-native compiler.
-Cobrust shipped its `0.1.0` stable tag on 2026-05-10 after a 9-week run
+Cobrust shipped its `0.1.0-beta` tag on 2026-05-10 and `0.1.0` stable on 2026-05-11, after an 11-day intensive run (first commit 2026-04-30 → v0.1.0 stable tag 2026-05-11; v0.1.2 + α Phase F.2 followed)
 with multiple parallel Claude agents (Opus 4.7 and Sonnet 4.6) coordinated
 via the methodology you'll find in [`SKILL.md`](./plugins/adsd/skills/agent-driven-development/SKILL.md).
 
@@ -126,7 +151,7 @@ project so the methodology can be tested outside its founding context.
 File an issue describing your project if interested.
 
 ADSD is **battle-tested but not orthodoxy**. Adapt it. If you find a
-failure mode we missed, propose F22+ via a PR to
+failure mode we missed, propose F31+ via a PR to
 [`reference/failure-modes-catalogue.md`](./plugins/adsd/skills/agent-driven-development/reference/failure-modes-catalogue.md).
 
 ## Contributing

docs/agent/conventions.md

@@ -0,0 +1,185 @@
+---
+name: ADSD repo conventions
+description: Meta-conventions for this repo itself. ADSD codifies how to manage AI-agent projects; this file applies ADSD discipline to the ADSD methodology itself. Agents contributing to this repo should read this first.
+type: convention
+version: 1.0.0
+date: 2026-05-12
+status: active
+relates_to: [SKILL.md §"Documentation Discipline", README.md §Contributing, CONTRIBUTING.md]
+---
+
+# ADSD repo conventions
+
+> ADSD is a methodology for managing AI-agent software projects. This repo IS such a project. Therefore, **ADSD applies to ADSD**. This file captures the meta-conventions specific to this repo's contributors (humans and AI agents).
+
+## Repo structure (binding)
+
+```
+agent-driven-development/
+├── .claude-plugin/marketplace.json     # Plugin marketplace catalog
+├── plugins/
+│   └── adsd/
+│       └── skills/
+│           └── agent-driven-development/   # The skill — auto-loaded by Claude Code
+│               ├── SKILL.md                # Main methodology (~36 KB)
+│               ├── reference/              # Deep-dive references (F-patterns, evals, prompts, etc.)
+│               ├── case-study/             # Founding case study (Cobrust N=1)
+│               └── templates/              # Templates for ADR / finding / dispatch / snapshot / handoff
+├── docs/
+│   ├── human/
+│   │   ├── zh/                             # Chinese user docs — 与 en 一一对应
+│   │   └── en/                             # English user docs — 1:1 parity with zh
+│   └── agent/
+│       ├── conventions.md                  # This file
+│       ├── adr/                            # Meta-ADRs for ADSD itself
+│       └── findings/                       # Findings about ADSD's evolution
+├── scripts/
+│   └── doc-coverage.sh                     # Enforces zh+en parity per ADSD §3 mandate
+├── CONTRIBUTING.md                         # Human-facing contribution guide
+├── LICENSE-APACHE / LICENSE-MIT
+└── README.md                               # Entry point
+```
+
+**Binding constraints**:
+
+1. Every file in `docs/human/zh/` MUST have a parallel file at `docs/human/en/` with the same filename. Enforced by `scripts/doc-coverage.sh`.
+2. Every reference under `plugins/adsd/skills/agent-driven-development/reference/` MUST have YAML frontmatter (`name`, `description`, `type`, `version`, `date`, `status`, `relates_to`).
+3. The SKILL.md `description` field is the auto-activation trigger — keep it keyword-dense and specific.
+4. ADRs in `docs/agent/adr/` are zero-padded sequential (`0001-*.md`, `0002-*.md`, ...). Once accepted, an ADR is immutable; supersede via a new ADR.
+
+## Frontmatter contracts
+
+### Reference files (in `plugins/adsd/skills/agent-driven-development/reference/`)
+
+```yaml
+---
+name: <Reference title>
+description: <One-line trigger / summary>
+type: reference
+version: <semver>
+date: <ISO date of last substantive edit>
+status: active | deprecated | candidate
+relates_to: [skill:SKILL.md §section, reference:other-file.md, ...]
+---
+```
+
+### Meta-ADRs (in `docs/agent/adr/`)
+
+```yaml
+---
+doc_kind: adr
+adr_id: <NNNN, zero-padded>
+title: <ADR title>
+status: proposed | accepted | superseded | deprecated
+date: <YYYY-MM-DD>
+last_verified_commit: <SHA or TBD>
+supersedes: [<adr_id>, ...]
+superseded_by: [<adr_id>, ...]
+relates_to: [<adr_id>, <finding-slug>, ...]
+---
+```
+
+### Findings (in `docs/agent/findings/`)
+
+```yaml
+---
+doc_kind: finding
+finding_id: <slug>
+last_verified_commit: <SHA>
+status: open | closed | partial
+discovered_by: <agent role + session ID>
+dependencies: [adr:<NNNN>, finding:<slug>, ...]
+---
+```
+
+## Bilingual docs mandate (ADSD §3 dogfood)
+
+The skill's SKILL.md §3 mandates that every public item gets entries in:
+
+- `docs/human/zh/<topic>.md`
+- `docs/human/en/<topic>.md` (1:1 parity)
+- Agent-facing schema (in this repo: SKILL.md + reference/)
+
+This rule applies to ADSD itself. `scripts/doc-coverage.sh` enforces zh+en parity.
+
+**Operative checks** (run by `doc-coverage.sh`):
+
+1. Every `docs/human/zh/*.md` has a parallel `docs/human/en/*.md`
+2. Every `docs/human/en/*.md` has a parallel `docs/human/zh/*.md`
+3. Parallel files have identical filenames (case-sensitive)
+4. (Future) Section headers are 1:1 between zh and en
+
+CI fails if any check fails.
+
+## When to add a new ADR vs amend SKILL.md vs add a finding
+
+| Change type | Where | Trigger |
+|---|---|---|
+| New methodology rule | `docs/agent/adr/NNNN-<slug>.md` | The rule affects ≥2 reference files, templates, or SKILL.md sections |
+| Refine existing reference | edit the reference file directly + note in commit | Single-file refinement |
+| Document an ADSD evolution event | `docs/agent/findings/<slug>.md` | Real-world ADSD use surfaced a gap or worked unexpectedly well |
+| Update SKILL.md | edit SKILL.md + cross-reference an ADR if it's a binding rule | Adds a new "Part N" or modifies an existing one |
+| New cross-pollination ref (Anthropic / OpenAI / other) | `plugins/.../reference/<slug>.md` | New industry pattern worth adopting |
+
+## When NOT to add an ADR
+
+- Bug fix in a reference doc (typo, broken link)
+- Updating frontmatter date / last_verified
+- Adding an example to an existing section
+- Re-organizing within a single file
+- Translation update (zh ⟵→ en sync)
+
+Per ADSD §"ADR vs Finding distinction": ADRs are forward-looking decisions; small refinements don't need them.
+
+## Commit message format
+
+```
+<type>(<scope>): <short description> [vX.Y.Z]
+```
+
+- `<type>`: `feat`, `docs`, `fix`, `refactor`, `chore`
+- `<scope>`: `skill`, `reference`, `case-study`, `templates`, `docs-zh`, `docs-en`, `meta`
+- Include `[vX.Y.Z]` semver if the change is release-worthy
+
+Examples:
+
+```
+feat(reference): add evals-first-development.md (v1.2.0)
+docs(zh): translate getting-started.md to match en parity
+fix(skill): correct cross-reference path after plugin layout migration
+chore(meta): bump SKILL.md description for trigger keyword coverage
+```
+
+Sign with session ID per F21 (Cross-session identity overload):
+
+```
+Co-Authored-By: Claude Opus 4.7 (session XYZ) <noreply@anthropic.com>
+```
+
+## Identity hygiene (F21 closure)
+
+Per F21 codification:
+
+- Do NOT sign as bare "review-claude" or "ADSD-author" in commits or files
+- Use session-stamped attribution: `review-claude (session 4bb35f43)` or `ADSD-author (session XYZ)`
+- Reserve plain handles for the abstract role in narrative prose only
+
+## Versioning policy
+
+- **v1.0.x** — initial release, plugin migration
+- **v1.1.x** — F19/F20/F21 codification
+- **v1.2.x** — cross-pollination references (Anthropic + OpenAI)
+- **v1.3.x** — bilingual docs + remaining G3/G5/G6/G8/G9/G10/G12 gaps
+
+Semver bumps follow SemVer 2.0:
+
+- MAJOR: breaking change to skill format, plugin layout, or canonical paths
+- MINOR: new reference file, new template, new ADR
+- PATCH: refinement, typo, frontmatter update, translation sync
+
+## Cross-references
+
+- `CONTRIBUTING.md` — human-facing contribution flow
+- `plugins/adsd/skills/agent-driven-development/SKILL.md` §"Documentation Discipline" — methodology origin of these rules
+- `plugins/adsd/skills/agent-driven-development/reference/failure-modes-catalogue.md` F1 family + F19 + F20 + F21 — the failure modes these conventions prevent
+- `scripts/doc-coverage.sh` — machine enforcement of zh+en parity