workflows-outbound-skills

10 Claude Code skills for running outbound end-to-end in your terminal. Built at Workflows.io after mapping our full GTM process, breaking it into standalone skills, and running them through hundreds of real-client campaigns.

Connect the right MCPs and Claude Code becomes your outbound team.

The 10 skills

#	Skill	What it does
1	fivos-icp-segmentation	Turn a client brief into an ICP matrix + tiered filter JSON
2	fivos-campaign-strategist	Pick plays, signals, sequence shapes, volume targets
3	fivos-account-sourcing	Source accounts from Discolike / AI Ark / Ocean.io with two-gate approval
4	fivos-account-enrichment	Enrich accounts through Clay (tech stack, funding, hiring, custom research)
5	fivos-contact-sourcing	Find decision-makers at target accounts via AI Ark + Clay
6	fivos-contact-enrichment	Email + phone waterfall via Findymail + BetterContact, Zerobounce verification
7	fivos-lead-scoring	Tier accounts A/B/C with a 10-row audit sample + reasoning
8	fivos-copywriter	Tier-aware email + LinkedIn sequences, AI-tell blocklist, no em dashes
9	fivos-campaign-launcher	Create + activate campaigns in Instantly + HeyReach
10	fivos-campaign-analysis	Weekly analytics + anomaly detection across sequencers

Quick install

git clone https://github.com/fivosaresti/workflows-outbound-skills.git ~/.claude/skills/fivos

Claude Code auto-discovers skills by reading the frontmatter in each SKILL.md. Nothing else to install.

Update later with cd ~/.claude/skills/fivos && git pull.

Configure API keys

Copy .env.example to your Claude Code env file (usually ~/.claude/.env) and fill in keys for whichever providers you're using. The skills degrade gracefully - they'll work with whichever providers you've wired and tell you what's missing for the rest.

# Account sourcing (pick any subset)
DISCOLIKE_API_KEY=
AI_ARK_API_KEY=
OCEAN_IO_API_KEY=

# Contact enrichment
FINDYMAIL_API_KEY=
BETTERCONTACT_API_KEY=
ZEROBOUNCE_API_KEY=

# Campaign launchers
INSTANTLY_API_KEY=
HEYREACH_API_KEY=

MCPs to wire

These MCPs are called by the skills. Install whichever ones you need - each skill degrades to a curl-based REST fallback if the MCP isn't available.

MCP	Used by	Status in v0.1
Discolike	account-sourcing	Expected prefix mcp__discolike__*
AI Ark	account-sourcing, contact-sourcing	Expected prefix mcp__ai_ark__*
Ocean.io	account-sourcing	Expected prefix mcp__ocean_io__*
Clay	account-enrichment, contact-sourcing	Expected prefix mcp__claude_ai_Clay__*
Findymail	contact-enrichment	Expected prefix mcp__findymail__*
BetterContact	contact-enrichment	Expected prefix mcp__bettercontact__*
Zerobounce	contact-enrichment	Expected prefix mcp__zerobounce__*
Instantly	campaign-launcher, campaign-analysis	Expected prefix mcp__instantly__*

HeyReach is REST-only (no MCP today) - uses HEYREACH_API_KEY directly via Bash + curl.

If your actual MCP tool names differ from these prefixes, a find-and-replace across the skills folder closes the gap. The skill logic doesn't change.

How the skills compose

The 10 skills form one pipeline. Run them in order per client / campaign:

icp-segmentation → campaign-strategist → account-sourcing → account-enrichment
                                                                 ↓
contact-sourcing → contact-enrichment → lead-scoring → copywriter → campaign-launcher → campaign-analysis

Each skill writes its output CSV to disk; the next skill reads it. The engineer stays in the loop at two approval gates: before spending on providers, and after the first 10-row qualification sample.

Hard rules baked in

These are enforced across the skills - don't fight them:

• Two approval gates on sourcing + scoring (pre-spend, post-sample).
• Contacts come from AI Ark + Clay only (we tried every other provider - this pair covers 95% of cases cheapest).
• AI Ark uses structured filters, never natural-language prompts - NL wastes credits.
• Stop on deliverable in the email waterfall. No "let's run one more provider to compare."
• No em dashes anywhere. Use -. The copywriter skill has an AI-tell blocklist and enforces this.
• Upload every qualified CSV to an internal DB (if you have one) tagged by client + campaign so past builds are reusable.

Internal company database (optional)

fivos-account-sourcing Step 0 queries an internal DB of companies you've already sourced. Skip if you don't have one. If you do, tag every upload by client, campaign, and a short contents description so past lists are filterable. The DB compounds - every campaign makes the next one cheaper.

Works with any DB: Postgres, BigQuery, Supabase, Snowflake. Update _query_internal_db.py / _upload_to_db.py in account-sourcing/ to talk to yours.

Verify install

Pop a new Claude Code session and run:

/skills

You should see all 10 fivos-* skills in the list.

End-to-end pilot

Once the MCPs are wired, run these in order against one real client:

1. fivos-icp-segmentation → ICP doc.
2. fivos-campaign-strategist → campaign plan.
3. fivos-account-sourcing - start with a 10-row sample.
4. fivos-account-enrichment on the sourced CSV.
5. fivos-lead-scoring - 10-row sample, audit reasoning, approve, full run.
6. fivos-contact-sourcing on qualified accounts.
7. fivos-contact-enrichment email waterfall.
8. fivos-copywriter 3 rows, review voice, full run.
9. fivos-campaign-launcher into a throwaway Instantly campaign (don't activate until you've reviewed).
10. fivos-campaign-analysis once live, weekly.

Known limitations in v0.1

• Lead scoring runs via Claude in-session - fast for <500 rows, slower at scale. Swap in a dedicated qualification model for big runs.
• HeyReach is REST-only; no MCP yet.
• Smartlead / Lemlist mentions in campaign-launcher are pattern-only - extend the skill yourself to activate them.
• No Apollo or HubSpot integration in v0.1.

Contributing

• Prompts, rubrics, provider-weighting updates: open a PR.
• Adding a new provider: add a sub-file under the relevant skill folder (e.g. account-sourcing/<provider>.md), document the MCP prefix + filter shape + pricing + strengths/weaknesses, then wire it into the main SKILL.md workflow.
• Bug or bad output: open an issue with the command, the input CSV shape, and what you got back.

Questions, feedback

Open an issue or DM me on LinkedIn: linkedin.com/in/fivosaresti.

License

MIT - see LICENSE.