kacperrucinski/personal_ai_infrastructure/contextsearch

ContextSearch

Search prior work for: $ARGUMENTS

Usage Modes

1. Standalone — Search, present findings, say: "Context loaded on [topic]. Most recent session: [X]. What would you like to do?"
2. Paired with request — Search first, then execute the request informed by context.

Phase 1 — Fast Index Scan (ALL IN PARALLEL)

Run all four searches simultaneously in a single response:

1A. Session Registry Read ~/.claude/PAI/MEMORY/STATE/work.json. Match entries where task, slug (object key), or sessionName contains "$ARGUMENTS" (case-insensitive). Extract: task, phase, progress, effort. Limit to 5 most recent matches.

1B. Session Names Read ~/.claude/PAI/MEMORY/STATE/session-names.json. Match entries where name contains "$ARGUMENTS" (case-insensitive). Extract: sessionId, name. Limit to 5 most recent.

1C. Work Directory Names

fd -t d -i "$ARGUMENTS" ~/.claude/PAI/MEMORY/WORK/ --max-depth 1 | tail -10

1D. ISA Title Grep Use Grep to search for "$ARGUMENTS" across ~/.claude/PAI/MEMORY/WORK/ with glob **/ISA.md. Use files_with_matches mode. Limit to 5 results.

Phase 2 — Conditional Deep Dive

SKIP Phase 2 entirely if Phase 1 returned 3+ matches. Phase 1 is usually sufficient.

Only if Phase 1 returned fewer than 3 total matches, run in parallel:

2A. PAI Git History

git -C ~/.claude log --oneline --all --grep="$ARGUMENTS" -i -10

2B. Current Project Git History

git log --oneline --all --grep="$ARGUMENTS" -i -10

After Search: Load ISA Summaries (NOT full ISAs)

From Phase 1C/1D matches, identify the 3 most recent ISAs (by directory date prefix). Read only the first 10 lines of each (title + summary). Do NOT read full ISAs — the main agent can request a full read if needed.

Output Format

Compact single list. Omit sections with no results. Keep total output under 40 lines.

═══ CONTEXT: $ARGUMENTS ═══════════════════════
 
📋 SESSIONS (newest first, max 5):
  - [slug] — [task] | [phase] | [progress]
 
📂 ISA SUMMARIES (max 3):
  - [dir name]: [first heading from ISA]
    Path: ~/.claude/PAI/MEMORY/WORK/[dir]/ISA.md
 
🔗 COMMITS (if Phase 2 ran):
  - [hash] [message] ([repo])
 
════════════════════════════════════════════════

After Results

Standalone: "Context loaded on [topic]. Most recent: [X]. What would you like to do?"

Paired: Proceed with the request. If deeper context is needed, Read the specific ISA path shown above.

Gotchas

• Searches PAI session registry, work directories, and ISAs — not published content (use _CONTENTSEARCH for that).
• Phase 1 (fast scan) may miss relevant sessions. If results seem incomplete, use phase 2 (full search).
• Session descriptions in work.json are AI-generated summaries. They may not capture every topic discussed.

Examples

Example 1: Find prior work

User: "what did we do with the Telegram bot?"
→ Phase 1: fast scan of session registry
→ Finds sessions: telegram-monitor-revival, fix-telegram-channels-plugin-broken
→ Returns session summaries with ISA links

Example 2: Resume previous work

User: "pick up where we left off on the feed system"
→ Searches work directories and ISAs
→ Finds latest ISA with phase/progress state
→ Provides context for cold-start recovery

Execution Log

After completing any workflow, append a single JSONL entry:

echo '{"ts":"'$(date -u +%Y-%m-%dT%H:%M:%SZ)'","skill":"ContextSearch","workflow":"WORKFLOW_USED","input":"8_WORD_SUMMARY","status":"ok|error","duration_s":SECONDS}' >> ~/.claude/PAI/MEMORY/SKILLS/execution.jsonl

Replace WORKFLOW_USED with the workflow executed, 8_WORD_SUMMARY with a brief input description, and SECONDS with approximate wall-clock time. Log status: "error" if the workflow failed.