Ir Index¶
Build or check the IR search index. Run 'build' to (re)encode dense vectors for indexed documents; builds across every IR source share one SQLite DB and serialize on a DB-wide advisory lock, so a build that arrives while another is running returns {skipped true, reason 'build_in_progress'} rather than queueing behind it. 'status' is never gated and returns per-source counts including dense_eligible_docs (how many docs CAN be encoded) and pending_eligible (real backlog — NOT doc_count vs vector_count, which is misleading because sources like conversation intentionally leave dense_text empty for tool-only spans).
Skill name: ir_index
Category: context
Parameters¶
| Name | Type | Required | Description |
|---|---|---|---|
action |
str |
No | 'build' (default) or 'status' |
days |
int |
No | Index sessions from last N days (default 30) |
force |
bool |
No | Rebuild from scratch (default False) |
source |
str |
No | Source to index: 'conversation' (default) |