Docs uplift plan (phased)¶
This page is the canonical “how we get to enterprise depth” plan for PenguiFlow docs.
It is intentionally procedural: it tells you what to upgrade, in what order, and what “done” means.
What “enterprise depth” means¶
For Tier A pages (the curated MkDocs site), the required rubric is defined in Docs style.
For Tier B pages (long-form internal guides), the bar is:
- accurate to the code
- clearly labeled as “source material” vs “canonical guidance”
- includes pointers to the Tier A pages where operators should start
For Tier C notes (RFCs/proposals), the bar is: labeled and coherent, but not blocking.
How work is tracked¶
- The complete file inventory and classification lives in Docs backlog.
- A PR should update:
- the Tier A pages it touches to meet the rubric, and
- any Tier B source pages it extracts from to add canonical pointers.
Tooling gates (must pass per PR)¶
uv run python scripts/check_md_links.pyuv run mkdocs build --strict
Optional (recommended once code snippets stabilize):
- add compile-only tests for the most important snippets (no network)
Phase 0 — Triage + backlog (1 PR)¶
Goal: make the work measurable and avoid “shallow but long” docs.
Deliverables:
docs/contributing/docs-style.mddefines Tier A rubric and conventionsdocs/contributing/docs-backlog.mdlists everydocs/**/*.mdfile with Tier/area/outcome/phase
Done when:
- every file under
docs/is listed in the backlog
Phase 1 — ReactPlanner deepening (2–3 PRs)¶
Goal: make planner docs production-ready for implementers and operators.
Primary sources:
REACT_PLANNER_INTEGRATION_GUIDE.md(Tier B)- planner contracts in
penguiflow/planner/react.pyandpenguiflow/planner/models.py
Tier A targets (must meet rubric):
- Planner overview (core contracts, dispatch/concurrency, output contract)
- Actions & schema (LLM-facing contract, repair/arg-fill failure modes)
- Tool design (typed schemas, side effects/idempotency, safe patterns)
- Parallel & joins (join injection sources, skip rules, runbooks)
- Pause/resume (HITL) (StateStore requirements, token handling, recovery)
- Memory (isolation, health states, persistence expectations)
- LLM clients (schema mode, streaming surfaces, provider gotchas)
- Observability + Troubleshooting
Done when:
- a new engineer can implement: catalog + tools + pause/resume + memory + parallel joins + event stream
- the planner pages include failure modes, monitoring guidance, and runnable patterns
Phase 2 — ToolNode + integrations (2 PRs)¶
Goal: external tools are production-safe on paper (limits, redaction, OAuth, persistence).
Primary sources:
docs/tools/*-guide.md(Tier B extraction sources)- ToolNode contracts in
penguiflow/tools/*
Tier A targets:
docs/tools/configuration.mddocs/tools/artifacts-and-resources.mddocs/tools/statestore.md- new pages: OAuth/HITL and MCP resources (when promoted into nav)
Done when:
- a platform team can deploy ToolNode-backed integrations with correct limits, redaction, OAuth, and durability
Phase 3 — Core runtime deepening (1–2 PRs)¶
Goal: runtime semantics are precise (envelopes, backpressure, cancel, retries).
Primary sources:
- runtime code in
penguiflow/core.py,penguiflow/node.py,penguiflow/types.py
Tier A targets:
docs/core/*docs/getting-started/*(kept correct and runnable)
Done when:
- engineers can reason about ordering/backpressure/retries/cancel without reading runtime code
Phase 4 — Operations runbooks (1–2 PRs)¶
Goal: deployment + observability docs are on-call ready.
Primary sources:
docs/deployment/worker-integration.md- observability primitives in
penguiflow/logging.py,penguiflow/metrics.py,penguiflow/middlewares.py
Tier A targets:
docs/deployment/production-deployment.mddocs/observability/*
Done when:
- operators can answer: “why is it slow”, “why is it stuck”, “what broke”, “can I replay”
Phase 5 — CLI + templates + contributor workflows (1 PR)¶
Goal: scaffolding and day-2 usage is obvious and correct.
Tier A targets:
docs/cli/*docs/contributing/*(dev setup, testing, releasing)
Done when:
- a team can bootstrap an agent project and use the playground/tools without tribal knowledge
Phase 6 — Legacy/manual consolidation (ongoing)¶
Goal: reduce duplication while keeping the curated site authoritative.
Primary sources:
manual.mdPENGUIFLOW_BEST_PRACTICES.md- architecture/spec/pattern notes under
docs/
Done when:
- no critical concept exists only in legacy long-form docs; a Tier A page exists and is linked