Skip to content

Guide

Reference

Concepts

index_events

Index the P6 event log at run granularity into the RAG infrastructure for operational intelligence queries (FP-0009 Component A).

Entry

scan

Final output

index_events_summary — indexed_runs, skipped_runs, filtered_runs, and new_cursor (ISO timestamp written to .reyn/index/events_cursor).

How it composes

Two-stage execution. Phase scan (LLM): a preprocessor runs resolve_scan_context to read the cursor file and summarise the event file inventory (count + timestamp range, no paths exposed); the LLM echoes the resolved since, event_files_count, skill_filter, and mode into scan_plan and finishes immediately. Skill.postprocessor (deterministic, no LLM): run_collect_chunks walks .reyn/events/ from the cursor, groups events by run boundary (1 run = 1 chunk), writes event_chunks.jsonl; the embed op embeds each chunk; index_write writes to the events source; run_advance_cursor persists the new cursor.

Caveats

  • Requires python (unsafe) permissions for resolve_scan_context, run_collect_chunks, and run_advance_cursor.
  • Incremental by default: only runs completed after the last cursor value are processed. Pass mode: replace to full-reindex.
  • Incomplete runs (no completion event yet) are skipped and deferred to the next invocation.
  • The source name events is hardcoded in the postprocessor's index_write step — this is an internal convention, not a P7 violation (the source name is not in OS code).

Usage

Use to make Reyn execution history searchable via recall queries. Run without arguments for incremental indexing; supply since or skills to narrow scope.

reyn run index_events
reyn run index_events --input '{"since":"2026-05-14T00:00:00"}'
reyn run index_events --input '{"skills":["my_skill"],"mode":"replace"}'

To recall indexed events from a skill or phase:

- op: recall
  query: "my_skill failure error phase"
  sources: ["events"]
  top_k: 10

Source

src/reyn/stdlib/skills/index_events/skill.md

On this page