| name | roast-case-study |
| description | Draft a LinkedIn-shaped before/after case-study card from a roast capture. Vlad-touch on every share — never auto-publishes. |
| metadata | {"clawdbot":{"emoji":"📝","tier":"strategy-c","handler":"roast_case_study"}} |
roast-case-study
Strategy C #2. Turns every roast capture into a shareable case-study card Vlad can paste to his LinkedIn personal feed. Vlad's LinkedIn audience (his existing brand) is the highest-trust traffic source for converting to customers, so each roast result becomes a public proof artifact instead of staying anonymous.
Pure drafter. Never auto-publishes. Vlad emoji-approves on Telegram, then pastes the card to LinkedIn personal himself.
Inputs (CLI)
python3 draft-case-study.py --domain <domain> --roast-summary "<summary>" [--dry-run]
python3 draft-case-study.py --lead-id rl_xxx [--dry-run]
--lead-id — load roast lead from local poll-state cache OR query https://api.meetrick.ai/api/v1/roast-leads/recent for the matching record
--domain + --roast-summary — use them directly (skip the lookup)
--dry-run — prints the card to stdout, writes nothing, sends no Telegram
Output
- Markdown file at
~/rick-vault/mailbox/drafts/case-study/<YYYY-MM-DD>-<sanitized-domain>.md with frontmatter:
---
kind: roast-case-study
target_channel: linkedin_personal
draft: true
review_required: true
domain: <domain>
created_at: <iso>
---
- Telegram alert posted to
customer topic via ~/clawd/scripts/tg-topic.sh:
📝 Roast case-study draft: <domain> — review at /draft N
So Vlad sees it land + uses /inbox UI to /draft N → /send N (he pastes manually to LinkedIn).
Card format (LinkedIn-friendly, ≤2500 chars)
🔥 Just roasted: <domain>
What I saw (60s scan):
- <pain point 1>
- <pain point 2>
- <pain point 3>
How I'd fix it:
1. <fix 1>
2. <fix 2>
3. <fix 3>
The honest version: most landing pages have the same 3 problems. Mine
too. The difference is whether the founder knows.
Want yours? https://meetrick.ai/roast — free, 60s, no email gate.
— Rick (autonomous AI CEO @ meetrick.ai)
Voice rules (non-negotiable)
- Founder-direct. First-person from Rick. Dry humor, opinion-first.
- Never name the prospect's email. Domain only — emails would feel creepy when Vlad shares this publicly.
- Real numbers only. $9 MRR / 1 paying customer (Newton). Don't fake it.
- Always end with the /roast CTA.
https://meetrick.ai/roast — free, 60s, no email gate.
- No filler openings. Banned: "Just spent some time looking at...", "I wanted to share...".
- Specific > generic. Pain points and fixes must reference what's actually wrong, not template-ish "your CTA could be stronger".
Safety / approval gates
- DRY-RUN by default — set
RICK_ROAST_CASE_STUDY_LIVE=1 to actually write the file + post to Telegram.
- Never auto-publishes to LinkedIn. There is no LinkedIn API code in this skill at all.
- All errors graceful no-op + log — never raises out of
main(). ~/rick-vault/operations/roast-case-study.jsonl.
- Vlad-touch on every share — Rick drafts, Vlad pastes. No exceptions.
Cost model
Single LLM call per draft via writing route (Sonnet). ~$0.05–0.15/draft. With 131 sessions/day capturing roast, even a 10% draft rate is ~13 cards/day → ~$1–2/day at steady state.
Bonus integration (deferred)
~/clawd/scripts/roast-lead-poll.py could subprocess-fire draft-case-study.py --lead-id <id> after each successful dispatch_event. Deferred to a follow-up commit — keep this ship clean and let Vlad opt-in by running the drafter manually first.