REMvisual/claude-handoff / skills/handoff
handoff
VerifiedCreate a structured session handoff when context is running low or work is pausing. Deep context mining, self-validation, multi-file splitting. Captures everything the next session needs.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 12.8 KB |
| references/close-session.md | markdown | 3.2 KB |
| references/mining-deep-chunked.md | markdown | 2.9 KB |
| references/output-template.md | markdown | 4.7 KB |
| references/validation.md | markdown | 2.1 KB |
references/mining-deep-chunked.md
View on GitHub# Deep and Chunked Mining Protocols
Used for sessions with >100K context tokens. Research shows LLMs have a "lost in the middle" problem — 30%+ accuracy drop for information in the middle of long contexts. A single-pass extraction at 500K+ tokens WILL miss decisions, measurements, and failed approaches from the middle ~50% of the conversation.
## Deep Pass (100K-500K tokens)
Two passes:
1. **Structured extraction** — Full checklist pass (see main skill). Force yourself: "Scan the MIDDLE third of the conversation for decisions and measurements I might skip."
2. **Gap-filling sweep** — Review your extraction. Ask: "What from the FIRST HALF is missing? What user feedback from MID-SESSION did I skip?"
## Chunked Pass (500K+ tokens, or 1M context + 50+ tool calls)
Map-reduce — a single pass will miss information at this scale.
1. **Segment** the conversation into 3-4 chronological chunks. Use natural breakpoints (topic shifts, major decisions, "let's move on" transitions).
2. **Per-chunk extraction** — Run the FULL extraction checklist against EACH chunk independently. Tag findings by chunk: `(early/mid/late)`.
3. **Merge + deduplicate** — Later decisions override earlier ones. Build a chronological timeline.
4. **Validation pass** — Ask: "What is missing for a new agent to continue? What comparison tables, cost data, or iteration histories did I skip?"
## Evidence Density Requirements
Chunked pass requires richer Evidence & Data. Heavy sessions produce commit logs, cost tables, approach comparisons, iteration histories, status matrices, and raw data. ALL must be captured — they are the most expensive to re-derive.
If your Evidence section has fewer than 3 tables or comparison data sets, you haven't mined deep enough.
## Line Targets
- **Chunked target:** 800 lines (1M context). Minimum: 500.
- **Deep target:** 600 lines. Minimum: 300.
A 10+ hour session with 100+ tool calls cannot be captured in 400 lines.
**Phase 1 baseline target (mandatory):**
- Deep: **300-400 lines on first write** — do NOT submit a 200-line Phase 1 intending Phase 2 to add the missing 100. That means you under-mined.
- Chunked: **500-600 lines on first write** — same rule.
Phase 2 (gap research + Edit) should push from baseline toward the ceiling (600 Deep / 800 Chunked). Phase 2 is for *additions* — tables you skipped, mid-session feedback you missed, raw data blocks you summarized. Phase 2 is not for "now let me actually fill in the sections I left thin."
If Phase 1 lands under its baseline, don't proceed to Phase 2 — rewrite Phase 1 first. The ceiling exists — USE IT.
## Anti-Skimming Rule
If you find yourself summarizing instead of extracting, STOP. Re-read the segment. The value of this skill lives in the specific details — numbers, file paths, function names, exact quotes from the user, error messages with line numbers. Abstract summaries ("improved performance", "fixed bugs") are worth nothing to the next session.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| slack-gif-creatoranthropics/skillsKnowledge and utilities for creating animated GIFs optimized for Slack. Provides constraints, validation tools, and animation concepts. Use when users request animated GIFs for Slack like "make me a GIF of X doing Y for Slack." | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| ponytailDietrichGebert/ponytailLazy senior dev mode for any coding task: the smallest change that fully solves it, and a reply a busy human understands in one read. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailHarvest every shortcut: comment into one debt ledger, so deferrals get tracked instead of forgotten. One-shot report. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailList every `shortcut:` comment (and older `ponytail:` ones) as a debt ledger. One-shot report, changes nothing. Use for "ponytail debt", "what did ponytail defer", "list the shortcuts", /ponytail-debt. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |