topprismwiki

Route TopprismWiki enterprise knowledge governance workflows for evidence-aware preview, formal update, query, context, graph, batch status, diagnostics, and source governance. Use when a user asks to

SKILL.md

GitHub

TopprismWiki

TopprismWiki is an evidence-governed knowledge layer for Agentic AI. This package exposes one entry skill; the files under subskills/ are internal modules and must not be installed or presented as peer Skills.

Routing

  • For approved chat or document updates, read the matching subskills/topprismwiki-formalize-*/SKILL.md file and use the deterministic runner in scripts/.
  • For formal knowledge queries, read subskills/topprismwiki-query/SKILL.md.
  • For bounded task context, read subskills/topprismwiki-context/SKILL.md.
  • For formal or candidate relationships, read subskills/topprismwiki-graph/SKILL.md.
  • For batches, watermarks, isolates, or health, read subskills/topprismwiki-status/SKILL.md.
  • For source approval, exclusion, readmission, retry, or validation, read subskills/topprismwiki-governance/SKILL.md.
  • For an error, blocked command, or unexplained batch, run diagnose and read the matching section of docs/zh-CN/troubleshooting.md or its English mirror.

Before routing, identify whether the request is read-only, a formal update, a source-governance action, or a diagnostic request. Read the capability matrix and do not present adapter-required or not-yet-supported behavior as built in.

Do not present a Workspace candidate, raw message, or document mention as a formal fact. Formal Vault changes require explicit update intent, a hash-bound review package, a dry-run, an atomic commit, and post-commit validation. The agent never edits the Vault directly outside the runner.

Global invariants:

  • Formal Vault is the default fact layer; Workspace evidence, candidates and raw messages must be labelled and never silently promoted.
  • No worker, model or child module writes vault/wiki directly. Formal changes require a reviewed, hash-bound package and the deterministic Runner.
  • Group names, member lists, filenames, co-occurrence and one-off brand mentions do not establish customer, project, delivery, adoption or person relationships.
  • Missing, ambiguous, conflicting or unprocessable evidence is isolated with a reason. Do not fill gaps with inference.
  • Never expose credentials, raw private evidence or unrestricted absolute paths in Markdown, logs, dashboard output or GitHub Issues.

Runtime

Run commands from the public package or an installed copy of this skill:

python3 scripts/topprismwiki.py doctor --capability all --project /path/to/wiki
python3 scripts/topprismwiki.py preview --source all --project /path/to/wiki
python3 scripts/topprismwiki.py update --source all --authorized --project /path/to/wiki
python3 scripts/topprismwiki.py query "project status" --project /path/to/wiki
python3 scripts/topprismwiki.py context "prepare a customer review" --budget 6000 --project /path/to/wiki
python3 scripts/topprismwiki.py graph "demo-project" --include-candidates --project /path/to/wiki
python3 scripts/topprismwiki.py batch list --project /path/to/wiki
python3 scripts/topprismwiki.py dashboard build --project /path/to/wiki
python3 scripts/topprismwiki.py diagnose --project /path/to/wiki

preview, query, context, graph, status, batch, dashboard, validate, and diagnose are read-only with respect to the formal Vault. retry and source governance update only private ledgers. update is the only command that may commit reviewed facts, and it requires --authorized.

Read references/architecture.md for data boundaries and references/dependencies-and-routing.md before operating a source-specific workflow. The examples under examples/ are synthetic and must not be replaced with real company data in a public repository.

For installation and the first runnable demo, read docs/en/getting-started.md or docs/zh-CN/getting-started.md. For an error, run diagnose and search the matching code in the bilingual troubleshooting guide.