An English coaching skill for Claude, built for working professionals who are already fluent enough to survive a meeting and stuck at "correct but shapeless."
The name is the lesson: in workplace English, the point goes first. Every framework here is a variation on that — lead with the conclusion, then support it.
The premise: their real problem is structure, not vocabulary. So the skill is built around five communication frameworks, and correction plays a supporting role.
The one rule — every session leaves a framework in the learner's hands. A session that ends with corrected sentences and no structural takeaway has failed, even if every correction was right.
| Framework | Use it for | Steps |
|---|---|---|
| TOPT | Presentations, slide by slide | Topic → Orientation → Point → Transition |
| P-A-R-A | Emails, status updates, Slack | Point → Action → Result → Ask |
| 3-to-1 | Storytelling, pitches, interview answers | Setup → Content → Punchline |
| Precision Checklist | Final review of anything high-stakes | Context → Facts → Interpretation → Conclusion/Action |
| CUT 50% | Anything bloated | Draft → Halve |
Each one ships with worked before/after examples, the failure mode learners actually hit, and a
drill. See references/frameworks.md.
Download point-first.skill from the
latest release and upload it in Claude under Settings → Capabilities
→ Skills.
Or clone this repo and point your Claude Code / agent setup at the point-first/
directory.
To build the .skill file yourself:
python -m scripts.package_skill point-first(from a checkout of Anthropic's skill-creator)
PF, let's practise. I'm a product manager and I run weekly sprint demos.
Or bring real material:
PF, apply TOPT to this slide.
PF, what should I have written instead? [paste email]
PF, run CUT 50% on this paragraph.
A full session runs about 15 minutes: warm-up, a five-item vocabulary batch, one framework drill against real work material, an open-ended challenge, then a score and a 48-hour task. Quick one-off questions skip the ceremony.
point-first/
├── SKILL.md # the coaching behaviour
└── references/
├── frameworks.md # the five frameworks in full
├── lexicon.md # 100 workplace verbs + 100 phrasal verbs, grouped
└── feedback.md # rubric, score weights, challenges, 48-hour tasks
A few decisions worth knowing before you fork it:
- No pronunciation scoring from text. Most sessions are text-only, and a pronunciation score invented from written words is worse than no score at all. Pronunciation enters the rubric only when the learner sends audio.
- Structure carries the heaviest weight alongside grammar. A learner with imperfect tenses and a clean P-A-R-A gets read; the reverse gets skimmed.
- Progress requires memory. The skill reads and writes a session log so streaks and trends are real. Without memory tools available, it says so instead of implying history it doesn't have.
- Scores are formative, never a credential. The skill corrects learners who treat them otherwise.
Aimed at intermediate-to-advanced non-native professionals — the lexicon and examples assume general office context. Beginners will drown; a specialist domain (legal, medical, aviation) needs its own lexicon file.
Built by Marcos Rezende — marcosrezende.com · hello@marcosrezende.com
Issues and pull requests are welcome. If you adapt the lexicon for a specialist domain, I'd like to hear about it.
MIT — see LICENSE.