ingest-llm-wiki-query

name	ingest-llm-wiki-query
description	Ingest a completed agent conversation into an LLM Wiki repository that uses queries, Knowledge Deltas, Summaries, Entities, index.md, and log.md. Use when a user explicitly asks to ingest or add the current query/conversation to the wiki, especially in repositories governed by lilpacy/CLAUDE.md, and optionally commit, push, or open a PR afterward.

Ingest an LLM Wiki Query

Establish the repository contract

1. Read the repository instructions, lilpacy/CLAUDE.md, lilpacy/index.md, and the relevant existing Summary, Concept, and Entity pages.
2. Read the current templates under lilpacy/schema/ before producing Delta, Summary, or Entity records.
3. Treat queries/ as immutable raw data after generation. Do not update Concepts during query ingest; record only Concept impact hints in the Delta.
4. Render generated Summary, Concept, and Entity math with Obsidian syntax: use $...$ inline and put each $$ block delimiter on its own line. Do not use \[...\] or \(...\) outside immutable transcripts or code examples.

Extract the transcript faithfully

1. Run ruby scripts/export_query_transcript.rb without arguments and identify the intended session by its displayed question and timestamp. Never select the latest session blindly.
2. Run the exporter with the exact session JSONL and a descriptive title.
3. Save the emitted Markdown without hand-editing its content. Freeze the transcript at that extraction point even if the live session later continues.
4. Preserve intentional Markdown trailing spaces in the generated transcript; exclude that immutable file from whitespace-only checks instead of rewriting it.

Build grounded wiki projections

1. Compute the transcript SHA-256. Name the Delta with the first 12 hex characters and record the full hash as source_snapshot.
2. Compare the transcript with the existing vault and record only new, refining, reinforcing, contradicting, or uncertain claims.
3. Copy every Delta evidence value as an exact contiguous transcript fragment.
4. Project every Delta summary_text verbatim into its declared required Summary section.
5. Update an Entity only for reusable facts or relations about an identifiable target. Point every Fact or Relation to the exact Delta block.
6. Register the Delta and Summary in index.md, keep the existing Entity catalog entry unless its description truly changed, and append a query entry to log.md.

Use the repository's deterministic ID algorithms when available. In the current pipeline:

• Delta ID: SHA-256 of source_link, classification, claim, and evidence joined by NUL; use the first 12 hex characters.
• Fact ID: SHA-256 of entity path, attribute, value, and Delta ID joined by NUL; use the first 12 hex characters.

Verify before committing

Run from the repository root:

ruby scripts/wiki_pipeline_lint.rb .
ruby -Itest test/wiki_pipeline_lint_test.rb
ruby -Itest test/pi_ingest_apply_test.rb
ruby -Itest test/export_query_transcript_test.rb

Do not pass lilpacy to wiki_pipeline_lint.rb; the script appends lilpacy internally, so that would inspect the wrong path.

Also verify programmatically that:

• the transcript SHA-256 equals the Delta source_snapshot;
• every Delta evidence fragment occurs exactly in the transcript;
• every summary_text occurs in its declared Summary section;
• every Entity evidence link resolves to an existing Delta block;
• only intended files are staged.

When the user requests commit or PR creation, follow the repository's commit and review skills, run the required post-commit read-only review, push only the task branch, and open the PR with verification results.