Decomposition of the project documentation (the public GitHub Pages docs site +
the durable in-repo reference/contributor docs) into independent, agent-distributable
work packages — one WP → one page → one agent → one PR, in the same spirit as
analysis/development-plan.md.
Status: decomposition + first authoring pass done. This plan decomposes the docs deliverable so it can be written in parallel. It is not gated on the Phase-0 GO verdict — the design/analysis source material already exists, so all conceptual documentation (Tracks O/A/S/C) was writable now, before any code, and has been written. The usage docs (Track U) are now also written — as design-target (pre-Phase-0) reference, each re-validated against real behavior as its build phase landed (as-built “How to read this page” banners, 2026-08).
Progress (2026-07-06):
- Enablers
DOC-P0(IA freeze, §2) andDOC-P3(style guide) — done.- Content fan-out
DOC-O1…O5,DOC-A1…A7,DOC-S1…S5,DOC-C1…C5— written (22 pages) via a 22-writer + 22-reviewer parallel Workflow, each traced to its sources. ADR setDOC-C4— 13 ADR files (CD-1…CD-10, LB1, LB2, docs-generator).- Usage
DOC-U1…U6— written as design-target (pre-Phase-0) via a second 6-writer + 6-reviewer Workflow (196–330 lines each), from the design/build-plan sources, each carrying a prominent pre-Phase-0 banner and marking undecided values(planned)/(leaning — unconfirmed). They no longer contradict the build; each is validated against real behavior when its phase lands (getting-started/config P1, hooks P2/3, dashboard/api P4, telegram P5).- Deferred (need the generator / live CI):
DOC-P1(choose+scaffold generator),DOC-P2≡WP-X7(Pages publish),DOC-P4(home/landing),DOC-P5(nav/theme/search).- Update (2026-08-25):
DOC-P2≡WP-X7is done —.github/workflows/pages.ymlpublishesdocs/on the stock GitHub Pages Jekyll builder and the site is live at https://ivanbbaev.github.io/agenthropic/;DOC-P1,DOC-P4andDOC-P5remain deferred (ADR-0013 still defers the generator choice).- Verified: 44 pages, 766 internal file-links, 0 broken; 76 anchor deep-links, 0 broken under the github-slugger algorithm (VitePress/Docusaurus dialect — the one class
DOC-P1/DOC-P5must re-validate against the chosen generator, since MkDocs slugs differently); every0.0.0.0is a labelled anti-pattern; no spawner/WebSocket-transport guidance; secrets are placeholders only.
Source of truth for content: docs/ai/DESIGN.md (design basis),
analysis/ (concept-analysis-v2, development-plan, external review,
animated-room), due-diligence/ (the evidence base), and the root
README.md / TODO.md / DONE.md.
DOC-P1). Generator-specific
frontmatter/components are added during assembly (DOC-P4/DOC-P5), not by content
authors.0.0.0.0, expose the port, or add a spawner. The
security posture (DESIGN §8) is a first-class, flagship doc, not an appendix.DASHBOARD_TOKEN=<token>), never real values.Physical content root is docs/site/ (published from docs/ by pages.yml; whether a
generator moves it is fixed by DOC-P1); slugs below are logical.
/ Home / overview (O1 → P4)
guide/
what-is-agenthropic Overview & the one-paragraph pitch (O1)
running Run book, written as-built (2026-08) (—)
the-moat Why build — the five absent features (O2)
comparison vs the field (baseline + 6 rivals) (O5)
roadmap Phases & waves, public-friendly (O3)
faq Self-hosted? cost? why not fork? (O4)
architecture/
overview Ingest loop + ports & adapters (A1)
data-model Schema, agent tree, token_usage, edges (A2)
hooks The twelve lifecycle events + catalog (A3)
ingest-reconciliation Source-of-truth, CD-1, JSONL vs hooks (A4)
dag-moat Persisted per-instance DAG + projection (A5)
cost-model Dual-pricing + delegation-savings (A6)
glossary Terms + hook-event reference tables (A7)
security/
model Loopback, token, no-spawner, no-SSRF (S1) ★ flagship
threat-model What the field got wrong; how we don't (S2)
remote-access Tunnel only (SSH / Tailscale) (S3)
operations/
backup-restore WAL, backup, tested restore, retention (S4)
troubleshooting Stuck-session watchdog, missing-Stop (S5)
contributing/
index Dev setup, PR flow, one-WP-one-agent (C1)
testing Fixtures, P0 tests, coverage >90% gate (C2)
licensing Clean-room rule + provenance scan (C3)
decisions/ ADRs: CD-1…CD-10, LB1/LB2 (C4)
governance SECURITY.md, CoC, issue/PR templates (C5)
usage/ ── WRITTEN pre-Phase-0, amended as-built (2026-08) ──
getting-started Prerequisites, install, run (U1)
hooks-installer Install the Claude Code hooks (U2)
configuration Env vars & config reference (U3)
dashboard The five daily questions + views (U4)
api Read API / SSE reference (U5)
telegram Alerts → @baev_bot_bot (U6)
Columns: WP · Sz (S≈½day / M≈1day / L≈2day of one agent) · Deps · Page · Scope & source inputs. Status is NOW unless a WP is tagged BLOCKED.
| WP | Sz | Deps | Page | Scope & source inputs |
|---|---|---|---|---|
| DOC-P0 | S | — | — | IA + site map sign-off. Freeze §2 above (sections, slugs, nav order). Gate for all content WPs so slugs/links are stable. |
| DOC-P1 | M | P0 | — | Choose & scaffold the generator (recommend VitePress — Vite-aligned with the leaning stack; alt: Docusaurus / MkDocs Material). Content root, local dev, build. Decision recorded as an ADR (feeds C4). |
| DOC-P2 | M | P1 | — | CI build & publish to GitHub Pages on merge to main. Public docs only; app stays 127.0.0.1. ≡ dev-plan WP-X7 — this fulfils it. |
| DOC-P3 | S | P0 | — | Docs style guide + page/ADR templates + terminology list. The convergence enabler for parallel authoring: heading structure, admonition conventions, code-sample rules (placeholder secrets), term casing (subagent, DAG, WAL). |
| DOC-P4 | M | P1, O1, O2 | / |
Home / landing page. Hero, the one-paragraph pitch, the five moat bullets, quick-links. Align with root README.md. |
| DOC-P5 | S | P1, P4 | — | Nav, sidebar, search, theme (light/dark), 404, edit-links. Final assembly polish once content lands. |
| WP | Sz | Deps | Page | Scope & source inputs |
|---|---|---|---|---|
| DOC-O1 | S | P3 | guide/what-is-agenthropic |
What agenthropic is; local-first, no telemetry egress. README.md; DESIGN §1. |
| DOC-O2 | M | P3 | guide/the-moat |
The five absent features & why greenfield. DESIGN §0, §2; concept-analysis-v2 §2. |
| DOC-O3 | M | P3 | guide/roadmap |
Phases 0–6 + the wave idea, public-friendly (no internal WP ids). DESIGN §9; development-plan §Phases/§Waves. |
| DOC-O4 | S | P3 | guide/faq |
Self-hosted? cost? privacy? why not fork simple10/hoangsonww? DESIGN §0, §8; due-diligence recommendation. |
| DOC-O5 | M | P3 | guide/comparison |
vs claude-code-templates baseline + the 6 rivals, public-friendly table. due-diligence market-landscape.md + recommendation.md. |
| WP | Sz | Deps | Page | Scope & source inputs |
|---|---|---|---|---|
| DOC-A1 | M | P3 | architecture/overview |
The ingest loop diagram + ports/adapters. DESIGN §3; concept-analysis-v2 (arch lens). |
| DOC-A2 | L | P3 | architecture/data-model |
agents(self-ref), sessions, events/events_raw, token_usage, orchestration_edges. Annotated DDL. DESIGN §4; development-plan Track D. |
| DOC-A3 | M | P3 | architecture/hooks |
The twelve lifecycle events, SubagentStart/Stop handling, what we do post-ingest. DESIGN §5. |
| DOC-A4 | L | P3 | architecture/ingest-reconciliation |
Immutable substrate + deterministic projection; CD-1 (JSONL-primary — empirically pre-answered CONDITIONAL-GO (conf 85) by the 2026-07-04 desktop probe; the formal Phase-0 spike confirms it on the paired-capture corpus); the Phase-0 probe. concept-analysis-v2 CD-1/LB1; analysis/phase0-probe.md; development-plan Track IN. |
| DOC-A5 | M | P3 | architecture/dag-moat |
Persisted, per-instance, dual-path orchestration_edges; rebuild-from-JSONL. DESIGN §2.1, §6; development-plan Phase 3. |
| DOC-A6 | M | P3 | architecture/cost-model |
token_usage buckets, compaction baselines, dual-pricing, delegation-savings. DESIGN §4 grafts; development-plan Track C. |
| DOC-A7 | S | P3 | architecture/glossary |
Glossary + hook-event reference tables (statuses, fields). DESIGN §4/§5; A2/A3 terms. |
| WP | Sz | Deps | Page | Scope & source inputs |
|---|---|---|---|---|
| DOC-S1 ★ | M | P3 | security/model |
Flagship. Loopback-only, mandatory timingSafeEqual token, no spawner, no SSRF, same-origin WS/SSE, no key in env. DESIGN §8. |
| DOC-S2 | M | P3 | security/threat-model |
What each rival got wrong (0.0.0.0, no-op token, /api/run RCE, SSRF) and how we structurally avoid it. DESIGN §8; due-diligence security.md. |
| DOC-S3 | S | P3 | security/remote-access |
Tunnel-only (SSH port-forward / Tailscale); never a reverse proxy to the open port. DESIGN §8. |
| DOC-S4 | M | P3 | operations/backup-restore |
SQLite WAL, backup + tested restore, retention/redaction. DESIGN §8; development-plan WP-F8, WP-D10. |
| DOC-S5 | S | P3 | operations/troubleshooting |
Stuck-session watchdog, missing-Stop→unknown, PreCompact repricing. (partial — deepens once code lands.) DESIGN §6; development-plan WP-IN12. |
| WP | Sz | Deps | Page | Scope & source inputs |
|---|---|---|---|---|
| DOC-C1 | M | P3 | contributing/index |
Dev setup, PR flow, the one-WP-one-agent model, the coverage >90% bar. development-plan §Global-DoD; CLAUDE.md. |
| DOC-C2 | M | P3 | contributing/testing |
Golden fixture corpus, the four P0 tests (three reconciliation proofs plus the five-daily-questions proof over real HTTP), the 12-scenario negative catalogue, merge-blocking coverage. development-plan Track X (X1–X5). |
| DOC-C3 | M | P3 | contributing/licensing |
Clean-room rule (cast/disler/nirdiamant all-rights-reserved), attribution (simple10/hoangsonww MIT), CI provenance scan. concept-analysis-v2 LB2/CD-9; development-plan WP-F5/F6. |
| DOC-C4 | L | P3, P1 | contributing/decisions/* |
ADR set: CD-1…CD-10 + LB1/LB2 + the generator choice (from P1), one file each, in a standard ADR template. concept-analysis-v2 §2–§3. |
| DOC-C5 | S | P3 | contributing/governance |
SECURITY.md (private-report path), Code of Conduct, issue/PR templates. Standard OSS governance + DESIGN §8 for the security-report policy. |
Written pre-code from the design/build-plan sources and since amended in place: each page
now opens with an as-built “How to read this page” banner (built / not built) where the
pre-Phase-0 banner stood, and marks any still-undecided value (planned) /
(leaning — unconfirmed). “Re-validate after” is the
phase whose real behavior the page must be checked against (and corrected if it drifted).
| WP | Sz | Deps | Page | Re-validate after |
|---|---|---|---|---|
| DOC-U1 | M | P3 | usage/getting-started |
Phase 1 (scaffold + server bootstrap WP-U0). |
| DOC-U2 | M | P3 | usage/hooks-installer |
Phase 2 hooks installer (WP-X8). |
| DOC-U3 | S | P3 | usage/configuration |
Phase 1 config surface (WP-U0). |
| DOC-U4 | M | P3 | usage/dashboard |
Phase 4 SPA + the five daily-question views (WP-U5…U9). |
| DOC-U5 | M | P3 | usage/api |
Phase 4 read API / SSE (WP-U1…U4). |
| DOC-U6 | S | P3 | usage/telegram |
Phase 5 Telegram sink (WP-A6). |
| Wave | What runs | WPs | Agents |
|---|---|---|---|
| D0 | IA freeze + authoring enablers (must precede content) | DOC-P0, DOC-P3 |
1–2 |
| D1 | The big fan-out — every writable-now content page, fully independent | DOC-O1…O5, DOC-A1…A7, DOC-S1…S5, DOC-C1…C5 |
up to 22 |
| D2 | Site infra + assembly (can overlap D1 — content is tool-agnostic) | DOC-P1, DOC-P2, DOC-P4, DOC-P5 |
2–4 |
| D3 | Usage docs — deferred; each fills when its build phase lands | DOC-U1…U6 |
as phases ship |
Critical path: DOC-P0 → DOC-P3 → {content} → DOC-P4 → DOC-P5 → DOC-P2 (publish).
The 22 content WPs in D1 have no inter-dependencies — that is the whole point of
the decomposition: hand each to its own agent.
To avoid double-tracking against analysis/development-plan.md:
| Dev-plan WP | Relationship |
|---|---|
WP-X7 (GitHub Pages build) |
Fulfilled by DOC-P2. Same deliverable. |
WP-X6 (README green badges + donation) |
Stays in the dev plan — done (TODO.md ticks it): the badges and the donation links have landed in the root README.md. DOC-P4 aligns the home page with the README. |
WP-X9 (RELEASE.md + per-role DoD) |
Stays in the dev plan (release track). Track C links to it. |
WP-X10 (WORKLOG discipline) |
Harness-local, git-excluded — not part of the public docs. |
WP-X11 (vector-DB EXPERIMENTAL stub) |
New surface this plan adds: the content of the docs site (Tracks O/A/S/C/U) — the
dev plan only carried the build/publish infra (WP-X7), never the pages themselves.
0.0.0.0 or spawn.DOC-P3 style guide (headings, admonitions, terminology).DOC-P1 lands); no broken-link warnings.WORKLOG.md entry appended for the WP.DOC-P1 / DOC-P5)Two things the pre-generator authoring cannot settle; both are verified-clean locally but need a decision when the generator lands:
docs/ai/. 18 docs/site/ pages cite ai/DESIGN.md as
their evidence source — as plain text, never as a hyperlink (2026-09-23 census: no Markdown
link targets it), so nothing dangles in the published site even though docs/ai/ is
git-excluded (harness-local). That is option (a) of the original three — (a) inline the
quoted spans and drop the hyperlink, (b) vendor the cited sections into a published
architecture/ page and repoint, or (c) publish a redacted DESIGN excerpt. Links into
docs/analysis/ are fine — that tree is tracked. (Do not commit docs/ai/ to fix this.)#anchor deep-links validate under the
github-slugger algorithm (VitePress/Docusaurus). MkDocs / python-markdown slug
differently (e.g. emphasis handling, stop-word stripping). If DOC-P1 picks a
non-github-slugger generator, re-run the anchor check against the chosen renderer.Plan of record for content decomposition. Design basis: docs/ai/DESIGN.md.
Build decomposition: analysis/development-plan.md.