New Skills! /handoff, /prototype, /review and /writing-* | Skills Changelog

Actionable Insights

1. Try a disposable handoff file before spawning another agent Use the repo shown in the video: mattpocock/skills. First experiment: create mktemp -t handoff-XXXXXX.md, ask the current agent to summarize only live decisions, open questions, artifacts, and recommended next skills, then start a fresh agent with that file. Benefit: less context rot; caution: never duplicate sensitive/raw artifacts into the handoff if a path reference is enough. Start by turning this into a small, reversible pilot: write down the exact input, expected output, owner, and success metric before changing the wider workflow. The useful detail from the analysis is: - mattpocock/skills: The public repo exists at github.com/mattpocock/skills; search snippets describe /grill-me and /grill-with-docs as alignment/planning skills. - 00:31-02:03 — GitHub handoff/SKILL.md is visible with mktemp -t handoff-XXXXXX.md, artifact references, and argument-tailoring instructions; this confirms the workflow mechanics. Treat the first run as an evaluation, not a migration: capture before/after examples, note where the method saves time or improves quality, and keep the old path available until the new one passes repeated checks. Watch for the main failure mode here: overgeneralizing the creator’s demo beyond the evidence. If the video or comments only showed a narrow case, keep the rollout narrow and require fresh proof before broad adoption.

2. Prototype unknowns as throwaway code, not discussion For UI uncertainty, ask for 3-4 variants behind ?variant= on the real route; for state/business logic, ask for a tiny terminal harness that exercises edge cases. Pass criterion: after 20 minutes, you can choose, reject, or combine options. Start by turning this into a small, reversible pilot: write down the exact input, expected output, owner, and success metric before changing the wider workflow. The useful detail from the analysis is: If a decision cannot be reasoned out, run /prototype: UI variants on a route or a logic harness for state transitions. If a decision cannot be reasoned out, run /prototype: UI variants on a route or a logic harness for state transitions. Treat the first run as an evaluation, not a migration: capture before/after examples, note where the method saves time or improves quality, and keep the old path available until the new one passes repeated checks. Watch for the main failure mode here: overgeneralizing the creator’s demo beyond the evidence. If the video or comments only showed a narrow case, keep the rollout narrow and require fresh proof before broad adoption.

3. Add a “do not implement yet” planning/grilling skill Wrap instruction blocks in XML-style tags such as <what-to-do> and <supporting-info> as Anthropic recommends for separating prompt sections. Evaluation: run the same plan twice and count premature implementation attempts. Start by turning this into a small, reversible pilot: write down the exact input, expected output, owner, and success metric before changing the wider workflow. The useful detail from the analysis is: Verdict: agree, high confidence.** Transcript shows a concrete problem: long planning/grilling sessions leave too little context for implementation. - mattpocock/skills: The public repo exists at github.com/mattpocock/skills; search snippets describe /grill-me and /grill-with-docs as alignment/planning skills. Treat the first run as an evaluation, not a migration: capture before/after examples, note where the method saves time or improves quality, and keep the old path available until the new one passes repeated checks. Watch for the main failure mode here: overgeneralizing the creator’s demo beyond the evidence. If the video or comments only showed a narrow case, keep the rollout narrow and require fresh proof before broad adoption.

4. Keep agent transitions artifact-aware Handoff should point to PRDs, ADRs, commits, issue URLs, and prototype directories instead of re-summarizing them. Checklist: current goal, decisions, blocked questions, files changed, tests run, next agent skill, stop conditions. Start by turning this into a small, reversible pilot: write down the exact input, expected output, owner, and success metric before changing the wider workflow. The useful detail from the analysis is: - Prototype/spike precedent: The idea aligns with long-standing software “spikes” and throwaway p - **Claim: handoff documents are useful for agent continuity. - Handoff is a manual, user-controlled alternative to black-box subagents: the user decides when to fork context and what survives. Treat the first run as an evaluation, not a migration: capture before/after examples, note where the method saves time or improves quality, and keep the old path available until the new one passes repeated checks. Watch for the main failure mode here: overgeneralizing the creator’s demo beyond the evidence. If the video or comments only showed a narrow case, keep the rollout narrow and require fresh proof before broad adoption.

5. Document skill transitions for teams Commenters explicitly asked for demos and “what skill comes next” examples. Add a workflow page: /grill-with-docs → /prototype → /handoff → implementation → /review. Start by turning this into a small, reversible pilot: write down the exact input, expected output, owner, and success metric before changing the wider workflow. The useful detail from the analysis is: - The strongest community signal is not “more skills”; it is clearer demos, workflow transitions, and examples that prevent skill sprawl. - The strongest community signal is not “more skills”; it is clearer demos, workflow transitions, and examples that prevent skill sprawl. Treat the first run as an evaluation, not a migration: capture before/after examples, note where the method saves time or improves quality, and keep the old path available until the new one passes repeated checks. Watch for the main failure mode here: overgeneralizing the creator’s demo beyond the evidence. If the video or comments only showed a narrow case, keep the rollout narrow and require fresh proof before broad adoption.

Core thesis

Matt Pocock argues that agent work improves when repeated collaboration patterns become explicit skills: handoff for context transfer, prototype for reducing design uncertainty, XML-tagged prompt sections to control instruction priority, and review/writing skills for repeatable quality loops.

Big ideas / key insights

• Handoff is a manual, user-controlled alternative to black-box subagents: the user decides when to fork context and what survives.
• Prototype treats code as research. The output is intentionally disposable; the decision it unlocks is the real artifact.
• XML-like structure is a prompt hygiene technique, not magic. It makes competing instruction blocks easier for Claude-like models to separate.
• The strongest community signal is not “more skills”; it is clearer demos, workflow transitions, and examples that prevent skill sprawl.

Best timestamped moments with interpretation

See the nested transcript page for the raw transcript. The moments below are selected interpretation points, not a transcript dump.

• 0:31-1:33 — Handoff skill definition: temporary file via mktemp, summarize current session, suggest next skills, avoid duplicating artifacts.
• 2:03-3:34 — Two patterns: fire-and-forget bug fix and DIY sub-agent handoff for prototyping/implementation.
• 4:04-6:09 — Prototype splits UI variants from logic/state harnesses; human taste remains the frontend evaluation loop.
• 6:39-7:40 — XML tags added to grill-with-docs to distinguish core task from supporting info.
• 8:10-8:40 — Issue labels changed from “needs triage” to “ready for agent triage,” clarifying automation handoff.

Practical takeaways / recommended workflow

1. Start with /grill-with-docs or equivalent questioning to resolve ambiguity.
2. If a decision cannot be reasoned out, run /prototype: UI variants on a route or a logic harness for state transitions.
3. Preserve the chosen prototype and decisions as paths.
4. Generate a handoff file for a fresh implementation agent.
5. Review against both spec fidelity and repo standards before merge.

Comment insights

Top comments reinforced three themes: (1) users love /grill-me and handoff-style continuity but want 1-2 minute demos for each skill; (2) practitioners already built similar /baton-pass, handoff, or open/close-shop flows, which supports the underlying need; (3) some users worry about too many skills and need transition examples rather than another giant catalog. A notable technical addition: commenters compare handoff with Codex/Claude conversation fork flows and point out cache/context-preservation tradeoffs.

Deep research

• mattpocock/skills: The public repo exists at github.com/mattpocock/skills; search snippets describe /grill-me and /grill-with-docs as alignment/planning skills.
• XML tags: Anthropic’s prompt-engineering docs explicitly cover XML structuring for Claude prompts, supporting the creator’s claim that section boundaries can improve clarity: Claude prompting best practices.
• Conversation fork vs handoff: Comments mention Codex/Claude fork-style flows. A fork preserves more raw context and can preserve cache; a handoff intentionally compresses and filters. The best choice depends on whether noise or lost detail is the bigger risk.
• Prototype/spike precedent: The idea aligns with long-standing software “spikes” and throwaway prototypes: reduce uncertainty before committing architecture. The AI-specific addition is rapid generation of multiple UI/state variants.
• Contradicting evidence/cautions: The video itself admits the XML fix is vibe-based rather than eval-backed. Skill repos can also become another prompt surface that bloats context or creates conflicting instructions if teams do not prune and test them.

Verdict

• Claim: handoff documents are useful for agent continuity. Verdict: agree, high confidence. Transcript shows a concrete problem: long planning/grilling sessions leave too little context for implementation. Practical takeaway: use handoff when a new focused context is more valuable than preserving every token.
• Claim: handoff is better than built-in fork/subagent flows. Verdict: mixed, medium confidence. It is more controllable and compact, but comments correctly note forks may preserve cache and full context. Use handoff for distilled continuity; use fork when exact hidden conversation state matters.
• Claim: prototyping is critical for AI engineering and frontend quality. Verdict: agree, high confidence. The UI frames show variant-based exploration; this directly addresses the common failure where agents cannot judge taste without human feedback.
• Claim: XML tags reduce prompt loudness. Verdict: agree directionally, medium confidence. Anthropic docs support XML structuring, but this specific skill change lacks measured evals. Treat as a low-cost prompt-engineering pattern, not proof of correctness.
• Claim: more skills will help everyone. Verdict: mixed, medium confidence. Comments show strong adoption, but also fatigue. Skills need demos, naming discipline, and pruning.

Screen-level insights

• 00:00 — Star-history chart shows mattpocock/skills cresting roughly 70k GitHub stars, validating that this is a widely watched workflow pattern, not a private prompt hack.
• 00:31-02:03 — GitHub handoff/SKILL.md is visible with mktemp -t handoff-XXXXXX.md, artifact references, and argument-tailoring instructions; this confirms the workflow mechanics.
• 04:04-06:09 — prototype/SKILL.md, LOGIC.md, and UI.md appear on GitHub. The UI path recommends route-local variants and a selector, making the visual design loop concrete.
• 06:39-07:40 — grill-with-docs shows XML-style prompt blocks. The visual step matters because it reveals the exact prompt-structure change behind the claimed behavior fix.

My read / why it matters

This is a useful systems-design video for agent operators: the durable insight is to turn repeated human-agent collaboration moves into named, testable workflow primitives. The risk is cargo-culting the skill names without adding evals, demos, or pruning.

Verification notes

• Source/evidence audit: Checked the extracted transcript/comment packet under youtube-extract/DNqsMXH6Eog/, visual frame metadata, and external web sources named above. Where official docs were unavailable or search results were secondary, the analysis labels uncertainty instead of treating the claim as settled.
• Transcript/comment/frame fidelity audit: Timestamp claims are tied to nearby transcript chunks and the key-frame paths captured by the processor. Comment insights are distilled from top extracted comments, not invented audience sentiment.
• Hallucination/overclaim audit: Verdicts separate confirmed facts, creator interpretation, and practical risk. Any pricing/performance/future-roadmap claims that depend on vendor behavior are marked mixed or uncertain.
• Actionable Insights audit: The top section was checked for concrete first steps, tools/commands/links, evaluation criteria, and cautions. Generic advice was removed in favor of workflow-ready bullets.
• Residual uncertainty: YouTube extraction can omit later comments; web search results may lag vendor changes. Re-check linked vendor docs before spending money, migrating production systems, or changing compliance/security posture.
• Actionable Insights audit: expanded to the newer detailed format with fuller implementation notes, evaluation checks, and cautions where the existing evidence supports elaboration.