last365days

name	last365days
description	Persistent long-term research tracker that builds dated Markdown timelines for topics/people. Use when user says "track this over time", "research timeline", "last365", "persistent profile", "save research history", or requests multi-session tracking on a topic. Do NOT use for one-off quick searches without persistence.
metadata	{"author":"zaydk","version":"1.4.1","upstream":"https://github.com/zaydk/last365days","compatibility":"Requires Python 3.10+. Uses last30days.py as research engine."}

last365days: Persistent Research Tracker

Deep research with persistent storage — each topic gets a dated Markdown profile that grows over time. Builds a massive timeline showing what's new since you last checked.

Output Location

Files saved to: ~/Desktop/last365days/ or $LAST365DAYS_OUTPUT_DIR

Quick Reference

Task	Command Pattern
Research + Save	last365days <TOPIC>
List Profiles	persist.py list
Read History	persist.py history <slug>
Match Topic	persist.py match "<topic>"

Reference Navigation

Load only when needed — reference files contain detailed documentation for specific use cases.

Reference	Load When
references/file-format.md	Debugging profile format issues or same-day deduplication problems
references/operations.md	Advanced workflows: browse, diff, export operations

Core workflow and all essential commands are in this SKILL.md.

Workflow: Iterative Research with Quality Gates

This skill uses iterative refinement — research quality improves with validation loops.

Phase 1: Context Gathering

Goal: Load prior history (if exists) to avoid duplication

Step 1: Parse Intent

• If no topic → persist.py list → stop
• If topic → persist.py match "<topic>"

Match	Action
exact/high	Load history → proceed to Phase 2
medium	Ask user: append to <slug> or create new?
none	Create new profile → Phase 2

Validation Gate: Confirm profile selection with user before proceeding

Phase 2: Research Execution

Goal: Gather comprehensive data

Step 2: Resolve X Handle (if applicable)

• Quick WebSearch for {TOPIC} twitter x.com
• If verified handle found, pass --x-handle=<handle>

Step 3: Run last30days Engine

LAST30DAYS_OUTPUT_DIR="${LAST365DAYS_OUTPUT_DIR:-$HOME/.local/share/last365days/out}" \
python3 "${CLAUDE_SKILL_DIR}/scripts/last30days.py" "<TOPIC>" \
  --emit=compact --no-native-web \
  ${X_HANDLE:+--x-handle="$X_HANDLE"} \
  ${DAYS:+--days="$DAYS"} \
  ${QUICK:+--quick} \
  ${DEEP:+--deep}

Validation Gate: Check exit code 0, output not empty Rollback: If script fails, run with --quick flag; if still fails, report error and stop

Step 4: WebSearch Supplement Target: recommendations, news, announcements, release Exclude: x.com, reddit.com (already covered)

Phase 3: Quality Check Loop

Goal: Ensure synthesis meets quality threshold

Initial Quality Criteria:

• At least 3 distinct findings
• At least 2 cross-platform patterns
• Source diversity (not all from same site)
• Novel information (not duplicate of prior history)

If quality < threshold:

1. Identify gaps ("only 2 findings, need more")
2. Run targeted WebSearch on missing angles
3. Re-synthesize
4. Re-validate
5. Max 3 iterations — stop if quality doesn't improve

Phase 4: Synthesis & Presentation

Structure:

1. What changed since {last_date} (if prior history)
2. What I learned (3-6 findings)
3. Key patterns (2-4 cross-platform themes)
4. Source stats (auto-read from report.json)

Phase 5: Persist Results

Goal: Save to profile with metadata

cat << 'SYNTHESIS_EOF' | python3 ${CLAUDE_SKILL_DIR}/scripts/persist.py append "<SLUG>" --title "<Display Name>"
<YOUR SYNTHESIS HERE>

## Key Findings
1. ...
2. ...

## Patterns
1. ...
2. ...
SYNTHESIS_EOF

Validation Gate: Verify file was written (persist.py read <slug>) Rollback: If persist fails, output synthesis to user anyway with note

Examples

Track a company over time

User: research Anthropic and save it

# Step 1: Match topic
python3 ${CLAUDE_SKILL_DIR}/scripts/persist.py match "Anthropic"
# Result: exact match found → loading Anthropic.md history

# Step 2: Run research with last30days
LAST30DAYS_OUTPUT_DIR="${LAST365DAYS_OUTPUT_DIR:-$HOME/.local/share/last365days/out}" \
python3 "${CLAUDE_SKILL_DIR}/scripts/last30days.py" "Anthropic" --emit=compact --no-native-web

# Step 3: Synthesize and append
cat << 'EOF' | python3 ${CLAUDE_SKILL_DIR}/scripts/persist.py append "anthropic" --title "Anthropic"
## 2025-04-07 Research Update

### What Changed Since 2025-03-15
- New: Claude 3.7 Sonnet released with extended thinking
- Update: Computer use API now generally available

### Key Findings
1. Launched "research mode" with 64k token extended thinking
2. MCP protocol adoption accelerating across AI tools
3. New pricing tiers for high-volume API users

### Patterns
- Release cycle accelerating (quarterly major updates)
- Developer tooling prioritized over consumer features
EOF

# Output to user: "Updated anthropic.md — 4 new findings since March 15"

Research a person

User: track research on Simon Willison

# Match: "simon-willison" not found → creating new profile

# Run research
python3 "${CLAUDE_SKILL_DIR}/scripts/last30days.py" "Simon Willison" --emit=compact

# Create profile
cat << 'EOF' | python3 ${CLAUDE_SKILL_DIR}/scripts/persist.py append "simon-willison" --title "Simon Willison"
## 2025-04-07 — First Research

### Key Findings
1. Released datasette-lite 0.3 with WASM improvements
2. Blogging about LLM tokenization edge cases
3. New tool: llm-fragments for context window management

### Patterns
- Consistent weekly blog posts on niche technical topics
- Tools focused on data exploration and LLM interaction
EOF

# Output: "Created simon-willison.md — will track updates over time"

List what you're tracking

User: what topics am I tracking?

python3 ${CLAUDE_SKILL_DIR}/scripts/persist.py list

# Output:
# 1. anthropic — Last updated: 2025-04-07 (5 entries total)
# 2. simon-willison — Last updated: 2025-04-07 (1 entry)
# 3. openai — Last updated: 2025-03-28 (3 entries total)

Troubleshooting

Issue	Cause	Fix
No such file	persist.py not found	Check ${CLAUDE_SKILL_DIR} is set
report.json not found	last30days failed	Check script output, retry
Same-day duplicate	Running twice in one day	persist.py auto-deduplicates same-day entries
Empty synthesis	No new findings	Report "No significant changes since last check"

Next Steps Pattern

After presenting, offer 2-3 specific actions:

1. Suggest comparing changes since earliest date (if history exists)
2. Recommend related topics to track
3. Offer to set up monitoring/alerts