@hy-sde-org/dsh-tool-openwiki
Model-facing repository wiki lifecycle tools — openwiki_begin,
openwiki_submit_plan, openwiki_next_page, openwiki_submit_page,
openwiki_finish — that drive the deterministic openwiki 0.4 engine core
(@hy-sde-org/dsh-openwiki) in-process. The five-tool contract and every
model-facing description match upstream openwiki 0.4, so harness agents run
the same resumable, claim-grounded wiki generation with no external
openwiki CLI and with codebase-memory for structural discovery.
Surface
One Cordis agent-plane plugin (mounts as a preset or profile-patch row,
injects tools + systemPrompt, registers no service of its own):
openwiki_begin— start or resume a durable run (.run.json) over a Git repository root; returnsstatus=noopfor clean updates.openwiki_submit_plan— validate and durably persist the ordered PageJob queue; init requires/openwiki/quickstart.md, paths are normalized.openwiki_next_page— first pending job with existing Markdown + Claims.openwiki_submit_page— complete the current job by proving its complete Claim set against the written page (front matter repair then Claims resolution and durable verification).openwiki_finish— deterministic finalization: planned/abandoned deletions, Mermaid validation, wiki index sync, link validation, generated provenance, Claims finalization + manifest replacement, run metadata, and.run.jsonremoval.
Configuration
- id: tool-openwiki
name: '@hy-sde-org/dsh-tool-openwiki'
config:
host: harness # stable host identity recorded in run metadata
producerActor: harness # provenance actor for engine-owned finalizers
Agent preset
examples/agent-preset/ is a ready-to-copy DeepSeek
Harness agent preset: the openwiki:tools prompt section is registered by
this package, and pairing the row with @deepseek-ai/dsh-tool-codebase-memory
gives agents structural discovery on top of the forced lifecycle sequence.
License
MIT — see the repository LICENSE. Tool-level wrappers keep
the upstream openwiki protocol contract and model-facing descriptions
(openwiki 0.4.3, Copyright (c) 2026 langchain-ai, MIT).