Kinic Wiki Query Workflow
Goal
Answer questions from Kinic Wiki VFS databases through kinic-vfs-cli.
Use this workflow for read-only search, inspection, summarization, and answer-only tasks.
CLI Reference
- Use
kinic-vfs-cli for wiki.kinic.xyz URLs and raw database IDs.
- Run
kinic-vfs-cli --help for the command list.
- Run
kinic-vfs-cli <command> --help before using an unfamiliar command.
- Use
docs/CLI.md as the full CLI usage reference when working inside this repo.
Workflow
- Extract the database ID from
https://wiki.kinic.xyz/db/<database_id>/... or from the user prompt.
- Use the mainnet default canister for
wiki.kinic.xyz; pass --canister-id only for an explicitly non-default canister.
- Start with
status --json to confirm target and access.
- Prefer
query-context --namespace /Knowledge --json for normal questions. Answer from returned nodes and evidence, not from search hits alone.
- Use
search-remote for content recall and search-path-remote for path or basename recall.
- Use
list-nodes for inventory and prefix discovery; do not treat inventory output as final evidence.
- Use
query-sql for 2 or more known-path body reads or narrow candidate classification.
- Use
read-node --json for one final evidence body, exact-value confirmation, or mutation-adjacent etag capture.
- Use
source-evidence --json only when the exact /Knowledge/... path is known and source refs are needed.
- Use snapshot/export commands only for whole-scope CLI reads or trusted snapshot sync. Check the command help before use.
Broad Questions
- Build a candidate set before answering broad, list, comparison, classification, or corpus-synthesis questions.
- Search multiple terms: the raw user phrase, key nouns, aliases, title terms, and discovered domain terms.
- Start with
/Knowledge. If it is sparse or misses, inspect /Sources and root prefixes.
- Separate title/path matches from topic-term matches before synthesis.
- Group evidence by work, source, path, and prefix.
- Mark cross-work generalization as inference.
- Preserve exact spelling, paths, IDs, titles, dates, and URLs.
Rules
- Never run destructive or write commands in this skill.
- Do not parse
wiki.kinic.xyz client-rendered HTML when CLI access is available.
- Do not use Kinic Memory MCP for Kinic Wiki URLs or VFS database IDs.
- Do not answer from
status, list-nodes, search-remote, or search-path-remote alone.
- Escape SQL literals correctly and do not put untrusted free text into
query-sql.
- If no candidate remains after content search, path search, and inventory fallback, answer
insufficient evidence.
Output
- Cite paths actually read.
- Include source URL and source path only when present in read content or metadata.
- For exact-value questions, answer the exact value first, then cite the path.
- For broad synthesis, include scope, evidence groups, confirmed points, inference, and coverage limits.
Agent catalogs
Use index.md for navigation or reading order when useful, not as exhaustive inventory or evidence for an answer. An empty, missing, or stale catalog must not stop content/path search and actual inventory discovery. query_context continues to prioritize the index to preserve navigation for broad questions. If large catalogs or other role pages consume the context budget, narrow the namespace or follow up with search and direct article reads. Reader-facing overview.md is synthesis, not evidence for exact claims.
When inspecting a narrower Wiki scope, pass its prefix as --namespace (or API namespace). The public API defaults to /Memory; never interpret that default as a complete Knowledge inspection.