hydrate

Hydrate

Fast disambiguation index. Identify ambiguous terms in an ask, locate matching graph entries via semantic search, and return a curated shortlist of IDs. Do not read full bodies, summarise context, or write to the graph.

Preconditions

Access the PKB strictly through PKB MCP tools at summary or snippet detail only -- never read full node bodies, and never attempt direct filesystem access. If PKB tools fail, halt and report (halt-on-failure).

$ARGUMENTS

Workflow

1. Extract ambiguous terms: Identify 2-3 terms carrying ambiguity (project names, technical terms, shorthand references).
2. Search with varied phrasing: Run 2-3 phrasing variants per term using PKB semantic search.
   • Semantic search indexes subtasks as their own documents, so no separate task-listing call is needed to avoid missing open work.
   • Budget approximately six calls; stop when two consecutive queries return no new results.
3. Filter to a shortlist: Retain only hits that directly affect caller planning. Discard generic vocabulary matches.
4. Emit flat shortlist: Group by kind (tasks, knowledge, terms), omitting empty kinds:

   - `<id>` -- <what it is, <=12 words> -- <relevance to ask>
   

   If a search yields nothing, state the exact queries and scope: no prior task (PKB search: '...', 0 hits).

Possible Overlap Flag

If an incomplete task appears to cover ground shared with the ask, lead with it at the top:

**Possible overlap** -- decide before creating anything new:

- `<id>` [<status>] -- <title> -- overlaps on <shared scope or touched files>