Search History — Structured Document Search

Search structured wrap-up output and field notes across eight document types. Complementary to claude-history (which searches raw conversation JSONL).

Step 1: Parse Query

Extract from the user's input:

• Keywords: the search terms (everything that isn't a flag)
• --type filter (optional): one of reports, issues, retros, memory, plans, research, decisions, evolution, or all. Default: all.
• --since YYYY-MM-DD (optional): only include files dated on or after this date. Applies to all dated file types (reports, issues, retros, memory, plans, research). Does not apply to Decision Log or System Evolution (single files, always searched in full). Default: no date filter.

If the input is just /search-history with no keywords, ask what to search for.

Step 2: Search Targets

Eight document types with their glob patterns:

Type Key	Label	Glob Pattern	Dated?
reports	Session Reports	Documents/Field-Notes/Logs/*-Session-Report*.md	Yes
issues	Session Issues	Documents/Field-Notes/Logs/*-Session-Issues*.md	Yes
retros	Session Retros	Documents/Field-Notes/Logs/*-Session-Retro*.md	Yes
memory	Memory Syntheses	Documents/Field-Notes/Logs/*-Memory-Synthesis*.md	Yes
plans	Plans	Documents/Field-Notes/Plans/*.md	Yes
research	Research Docs	Documents/Field-Notes/Research/*.md	Yes
decisions	Decision Log	Documents/Field-Notes/Decision-Log.md	No
evolution	System Evolution	Documents/System/System-Evolution.md	No

When --type is specified, search only that type. Otherwise search all eight.

Step 3: Execute Search

For each active document type:

1. Glob to find matching files.
2. For dated types (reports, issues, retros, memory, plans, research): if --since is set, filter filenames by date prefix (YYYY-MM-DD >= since date). Files without a parseable date prefix are included.
3. Grep for keywords across the matched files. Use case-insensitive search. Request content output mode with 2 lines of context (-C 2).
4. For Decision Log and System Evolution (single files): --since is NOT applied — always search the full file.

Run searches for independent document types in parallel where possible.

Step 4: Present Results

Count-first mode

If total matches exceed 10 across all types, present a count summary first:

Match counts:
  Session Reports: 4 matches (3 files)
  Plans: 8 matches (5 files)
  Research Docs: 2 matches (1 file)
  Decision Log: 1 match

Then ask: "Want to see all results, or drill into a specific type?"

If the user says "all" or the total is <= 10, show full results immediately.

Full results

Group results by document type. Within each group, sort newest-first (filename date prefix makes this natural).

Format each match as:

**<Document Type>**
`file_path:line_number` — <matching line with context>

Multi-session days: When multiple files share the same date prefix (e.g., 2026-03-18-Session-Report.md and 2026-03-18-Session-Report-3.md), note it: "3 reports for 2026-03-18."

Zero results

If no results found across all types:

1. Suggest broadening keywords or removing --type filter.
2. Offer the exact claude-history fallback command:

   claude-history -L
   

   Then search interactively for the keywords. This searches raw conversation transcripts, which may contain what the structured docs missed.

Step 5: Synthesis (optional)

After presenting matches (or after count-first drill-down), offer:

"Want me to read the top matches and summarize the timeline?"

If the user accepts, read the 3-5 most relevant files (highest match density or most recent), then write a 1-2 paragraph narrative synthesis: what happened, when, and how the topic evolved across sessions. This turns search-history from a search tool into a narrative reconstruction tool.

Step 6: Follow-Up

After presenting results, offer:

• "Want me to read the full file for any of these?"
• "Want to narrow the search with --type or --since?"
• "Want me to summarize the timeline across these matches?"

Disambiguation

This skill searches structured wrap-up output and field notes (session reports, plans, research docs, decisions, evolution). For searching raw conversation transcripts, use claude-history (Homebrew tap — claude-history -L, then search interactively).