This Has Ended Up Being Better Than Expected and Fills...  🔗 **Source:** [This Has Ended Up Being Better Than Expected and Fills...](https://twitter.com/mattpocockuk/status/2094156122441625770/?rw_tt_thread=True) > [!note]+ Document Notes > tags: [agentic-workflows, claude-code-skills, spec-driven-development, multi-agent-implementation, afk-agents, matt-pocock, subagent-orchestration, credit-routing] > author: Matt Pocock > source: X (formerly Twitter) — @mattpocockuk > reference: https://twitter.com/mattpocockuk/status/2094156122441625770/?rw_tt_thread=True > date created: 2026-09-03 12:40 > > <!-- deep-read: light-touch --> > > > [!info] Generated analysis — not the author's words. Sections below are machine-written > > from the captured source. Only text inside `> [!quote]` blocks is verbatim from the author. > > Do not quote or attribute any other line to the named author. > > # 🏷️ Title: This has ended up being better than expected and fills an interesting hole in my skills > > > [!note] Why this is a light-touch note > > **128 words** across two tweets: a 2026-08-30 update quoting Pocock's own 2026-08-21 announcement of an `/implement-spec` skill. No code, no repo link that resolved (one `t.co` short link, unexpanded in the capture), no results beyond "better than expected." The claims are a builder's progress note. Enough for two atoms and a comparison to Courtney's own setup; not enough for eighteen sections. > <!-- deep-read:sections start --> ## 📝 Summary Pocock announced (Aug 21) an `/implement-spec` skill for Claude Code: "a multi-agent implementer that takes in a spec and tickets, does codebase research in a subagent, implements all the tickets in subagents with maximum concurrency, reviews the final code against the spec, cleans up all worktrees," aiming to "smash out huge chunks of work autonomously with minimal supervision." Nine days later (Aug 30) he reports it "has ended up being better than expected," then ranks it below something else: "Most folks don't have an AFK agent workflow set up yet. You should. It's better than /implement-spec." The skill is still worth running "especially while tokens are cheap, and very simple to set up," and he'll publish it to his public skill set after a holiday. The interesting move is the self-demotion. The thing he built is good; the thing he's recommending is the away-from-keyboard workflow he hasn't described yet. `/implement-spec` is positioned as the on-ramp: cheap, simple, interactive. The AFK workflow is the destination. ## 🔖 Atoms ### [H1] Spec-in, tickets-out, subagents-per-ticket, review-against-spec, cleanup > [!quote] ID:rw_01m1b1kqcmxnc6p2md5e3x0q21 > "Essentially a multi-agent implementer that: Takes in a spec and tickets · Does codebase research in a subagent · Implements all the tickets in subagents with maximum concurrency · Reviews the final code against the spec · Cleans up all worktrees" Five stages, one skill. The shape is the same one Courtney's session ran today with deep-read forks: brief in, one subagent per unit, lead reviews, lead cleans up. The review step is the gate; "maximum concurrency" is the cost lever. Tags: `#atom/literature/h1` ### [H2] The builder ranks his own tool second > [!quote] ID:rw_01m1b1kqcmxnc6p2md5e3x0q21 > "Most folks don't have an AFK agent workflow set up yet. You should. It's better than /implement-spec." An interactive multi-agent skill is the stepping stone; the unattended workflow is the goal. Said by the person who built the stepping stone, nine days after shipping it. Tags: `#atom/literature/h2` ## 💡 Personal relevance - **She already has the plugin.** `mattpocock-skills` is installed in her Claude Code (tdd, prototype, research, domain-modeling, codebase-design, code-review, grilling, writing-for-agents). `/implement-spec` isn't in it yet; he says it lands after his holiday. Watch the plugin's changelog rather than the tweet. - **The five-stage shape is her fork pattern.** Today's deep-read batch (12 forks, one doc each, lead runs backlog/mirror/gate) is `/implement-spec` with documents instead of tickets. The difference is the review stage: hers is a script (`deepread_check.py`), his is "reviews the final code against the spec," which is a model judging a model. Her CLAUDE.md rule ("A finding is a claim until something runs") is the sharper version. - **"While tokens are cheap"** is the routing note. Her credit-routing skill says the same thing from the other side: mechanical implementation goes to subagents; the lead reviews. Pocock's skill is that policy as a command. - **The AFK workflow he's pointing at** is what her Workflow tool and `/loop` already do. When he publishes his version, compare the gate: does his unattended run end at a verification step or at "done"? ## ✏️ Fleeting Queue - Expand the `t.co/lTmPYXkUx7` link when convenient; it's probably a screenshot or the skill file. `#atom/fleeting/f1` - When `/implement-spec` ships in `mattpocock-skills`, diff its review step against `superpowers:verification-before-completion`. Two takes on the same gate. `#atom/fleeting/f1` ## → Things 3 (max 2 actions) 1. **Check `mattpocock-skills` for `/implement-spec` in two weeks** (post-holiday); if present, read the review stage before running it on anything real. `#action/h3` <!-- deep-read:sections end --> 📄 **Full text:** [[Full Document Contents/Tweets/This Has Ended Up Being Better Than Expected and Fills....md|This Has Ended Up Being Better Than Expected and Fills... by Matt Pocock]] ## Highlights - Most folks don't have an AFK agent workflow set up yet. You should. It's better than /implement-spec. ([View Highlight](https://read.readwise.io/read/01m1kr9wsdy2c6fkz82n0f7254)) - Note: H2 — The builder ranks his own tool second: the interactive skill is the on-ramp, the unattended workflow is the goal #atom/literature/h2 - Tags: atom/literature/h2 - Should be able to smash out huge chunks of work autonomously with minimal supervision. ([View Highlight](https://read.readwise.io/read/01m1kr9zwrq4ka8s1j6ktyd6gc)) - Note: H1 — Spec in, one subagent per ticket, review against spec, cleanup: same shape as the deep-read fork pattern; the review step is the gate #atom/literature/h1 - Tags: atom/literature/h1