What if the fastest coder in the room is making you dumber?
You’re knee-deep in a Claude Code session. Forty minutes gone. Files re-read. Decisions? Blurry. That’s not Claude’s fault. It’s yours for letting it sprint ahead without a map.
Enter FRAME. A clever little Markdown trick for Claude Code that slams on the brakes. No installs. Just copy files, type /frame load sw-development, and suddenly Claude’s got a role – Requirements Engineer – and a gate. It won’t touch code till you sign off.
Here’s the pitch from its creator: > “FRAME didn’t produce better code. It produced a documented contract for what the code was supposed to do — agreed before anything was written, verified after.”
Spot on. But let’s gut this thing.
What Even Is This FRAME Nonsense?
Cartridges. Phases. Gates. Sounds like a bad video game. But it’s pure Markdown genius for Claude Code.
Five phases: SHAPE, BREAKDOWN, DESIGN, BUILD, CHECK. SHAPE spits out a PROJECT.md – goal, stack, constraints, scope. No code yet. You approve or yell stop.
BREAKDOWN chops the mess into bite-sized units. DESIGN sketches interfaces. BUILD codes one unit at a time. CHECK reviews and tests. Gates everywhere. Context clears? State saves to file.
Tested on a bug: _get_daily_limit() crashes on bad env vars. Plain Claude Code? One minute fix. Perfect. FRAME? Ten minutes. Same fix, plus docs, plus edge-case tests plain Claude missed – like whitespace strings.
Slower. Structured. Sharper.
And here’s my hot take Claude’s creators won’t love: this exposes how Claude Code is basically a caffeinated intern – brilliant, but zero process. Anthropic hypes speed; FRAME calls bullshit and adds scaffolding. Reminds me of the 1970s structured programming wars. GOTO everywhere? Chaos. Dijkstra said no. FRAME’s saying no to AI free-for-alls.
Why Bother Slowing Down Claude Code?
Speed kills in software. You know it. I’ve shipped bugs because ‘it works’ skipped the ‘what’s it supposed to do?’ chat.
FRAME shines when specs blur. Ambiguous features. Stale codebases. Audits for bosses. Outputs? Not just code – a full audit trail. Hand PROJECT.md to a teammate; they grok it instantly.
Plain Claude Code: output. FRAME: output + record. That’s the win.
But is ten minutes ‘overhead’? Hell no. It’s insurance. Predict my bold call: in six months, every serious AI coding setup will ape this. Or regret it when stakeholders ask, “Why’d we build that?”
Look, Claude Code’s great for trivia fixes. But real work? Decisions lurk. FRAME drags ‘em out early.
Does FRAME Actually Fix Claude Code’s Chaos?
Same bug, twice. Plain: zip. FRAME: questions first. Branch? Constraints? Criteria?
Produces this SHAPE block:
Goal: Fix _get_daily_limit() crash - non-numeric or zero env var Stack: Python 3.9+, pytest Branch: fix/frame-daily-limit-validation Constraints: none Acceptance: - Non-numeric → returns 50, no exception - Zero → returns 50, no exception … (and more)
You tweak. Then code. QA flags gaps. Whitespace test added. Boom – robustness.
Plain Claude intuited edges. FRAME systematized ‘em. For solos? Meh. Teams? Gold.
Critic’s aside: Anthropic’s PR spins Claude as ‘agentic.’ Cute. But agents without rails? Toddlers with flamethrowers. FRAME’s the nanny.
When FRAME Saves Your Ass (And When It Doesn’t)
Right tasks: features with fuzzy edges. Code you forgot. Plans needing buy-in.
Tur-Tur effect – dev’s term for ‘big scary goal becomes checklist.’ Overwhelm vanishes.
Wrong tasks: trivial bugs. Known punches. There, Claude Code laps it.
I’ve tried it. Simple bug fix? Skip FRAME. Prototyping a dashboard? Load it up. Docs alone justify the drag.
One gripe – Markdown-only means Claude fidelity matters. Cursor or VS Code integrations? Tweak needed. But zero deps? Pure.
The Real FRAME Power: Audit Trails That Don’t Suck
Final output: archive of SHAPE, phases, decisions. Reconstruct sessions blind.
In audits, this sings. Stakeholder asks why? Point to file. No ‘uh, trust me.’
Historical parallel: remember CMMI? Process maturity models bored devs to tears. FRAME’s lightweight cousin – AI-native, zero overhead beyond minutes.
Bold prediction: Open source this fully (it’s Markdown, duh), and it’ll fork into every LLM coder. Claude, Cursor, whatever.
But here’s the rub – if you’re solo cowboy coding, FRAME feels fussy. Teams? Mandate it.
So, yeah. Slow your AI down. Or stay lost.
🧬 Related Insights
- Read more: Gemma 4 is Finally Open Source—Here’s What Actually Works
- Read more: CTOs Under Pressure: How Chief AI Officers Are Reshaping Tech Leadership
Frequently Asked Questions
What is FRAME for Claude Code?
FRAME’s Markdown cartridges that structure Claude Code sessions into phases like SHAPE and BUILD, with gates for approval. No installs needed.
Does FRAME make Claude Code slower?
Yes, deliberately – from 1 minute to 10 for bug fixes. But adds docs, edge cases, and audit trails worth it for complex work.
When should I use FRAME with Claude Code?
Ambiguous features, team handoffs, audits. Skip for quick, clear tasks.
Is FRAME open source?
It’s just Markdown files – copy, tweak, share freely.
