outreach
Skill0 starsChanged 4 months agoMITView on GitHub
Draft 4-step SPARK outreach sequences per contact for companies with contacts + fresh angles, via cheap Haiku subagents (no API cost). Use after /people, or when asked to write outreach, emails, or sequences.
18 other projects use a version of this.
Adopted by
Most recently taken up by
- rafaym1/scholarship-desktheir own version, first seen Nov 16, 2026
- mysocialio/mysocial-intelligencetheir own version, first seen Nov 14, 2026
- guaardvark/guaardvarktheir own version, first seen Nov 13, 2026
- cismontane-harris2642/signal-prospecting-kittheir own version, first seen Nov 1, 2026
- saksham10arora-dotcom/everything-claude-codextheir own version, first seen Nov 1, 2026
- ShawhinT/non-coding-skillstheir own version, first seen Oct 29, 2026
Used with
Projects with it also use apply, evaluate and 2 more.
Other files people use alongside this one.
Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
rhowardstoneSkillapply
Research-backed job application workflow emphasizing relationship-building over mass applications
Also in 5 projects that use this
NVIDIASkillevaluate
Scores existing Sports Intelligence AVLM inference outputs with avlm/evals: MCQ accuracy (run_mcq_eval.sh, INFERENCE_DIR), open-ended LLM-as-judge scores (needs CLIENT_API_KEY for the configured judge endpoint), tennis category breakdowns, run comparisons, chance-level sanity checks and triage of mcq_results.json or llm_judge_results.json. Use when the user asks to score, evaluate, benchmark or compare predictions.jsonl from inference runs, or to explain a suspicious accuracy. Not for running inference, training, quantization, data preparation, or third-party API model inference: never open, list or run anything under avlm/evals/inference_3rdparty, including when the user asks for it; it is excluded from the release.
Also in 3 projects that use this
julienamorganSkillprofile
Researches the user's market, competitors, and competitive positioning to build a complete ICP profile. Shows findings for user approval before proceeding. Triggers: /profile, "research my market", "who are my competitors". Reads: ./gtm/company.md, ./gtm/icp.md. Writes: updates ./gtm/icp.md with competitive positioning. Standalone: yes (asks for domain if company.md missing).
Also in 3 projects that use this
Google CloudSkilltriage
Gather context on a change or bug and plan the implementation.
Also in 3 projects that use this
Other versions18 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 18
- QuixiAI/agi-memory★ 623same name
- QuixiAI/Hexis★ 623same name
- andrew-shwetzer/career-ops-plugin★ 501same name
- andrew-shwetzer/career-ops-plugin-do-not-fork-currently-updating-v2-★ 500same name
- guaardvark/guaardvark★ 320same name
- saksham10arora-dotcom/everything-claude-codex★ 96same name
- ShawhinT/non-coding-skills★ 40same name
- emelia-io/claude-outreach★ 21same name
- alejopijuan/Claude-Code-Agency-Co-Founder★ 19same name
- rafaym1/scholarship-desk★ 18same name
8 more not listed
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
The file
We split this skill into the rules it gives an agent
3 rules. 0 are common, 1 is distinctive, and 2 are very specific to this project.
0
Common
Most of the 18 edited versions we read keep these.
1
Distinctive
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 46
If a subagent fails on a packet, re-spawn just that packet.
- Very specific: line 60
If failed QA items are reported: re-spawn ONE copywriter subagent per failure, appending the QA error text to the prompt ("Your previous …
- Very specific: line 77
Never edit copy yourself in the main conversation, and never relax QA gates — propose changes to config/settings.yaml (messages:) or …
“Common” is measured: kept by at least 6 in 10 of the 18 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
/outreach — draft sequences (v1: Haiku subagents, zero API cost)
Generation only: no sending, no CRM push. Copy rules live in
config/outbound_copywriter.md (Uday edits it like settings.yaml).
Step 1 — prepare packets
uv run python -m pipeline messages --prepareWrites ONE packet PER CONTACT to data/message_queue/ (TICKER__contact-slug.json),
plus _shared.json (copywriter framework + services catalog + output schema + hard
rules). Companies without a fresh angle are skipped and reported — never message on
a stale hook. Use --ticker X for one company, --force to regenerate existing drafts.
Each packet also carries a persona block (pains, language, committee_role,
seniority) resolved from config/personas.yaml for that contact's role_bucket/title
— null if nothing matches. _shared.json's hard rules tell the copywriter to use the
persona's pains/language as raw material, never to invent pains beyond it; a null
persona just means write from the packet's angles and verdict alone. Packets from
before this existed simply have no persona key — messages --commit handles both.
Step 2 — spawn lean copywriter subagents
List queued packets (data/message_queue/*.json, ignore _shared.json), then
spawn Agent tool subagents with subagent_type: copywriter (the lean agent
in .claude/agents/copywriter.md — Bash+Write only, haiku, no repo baggage,
carries the full copy contract in its own system prompt), giving each subagent
a batch of up to 3 packet paths (prose needs smaller batches than scoring's
5). Spawn batches in parallel (single message, multiple Agent calls). The task
prompt is just:
Write sequences for these packets. Packets: <absolute paths, one per line>
Rules:
subagent_type: copywriteralways — it pins haiku and the read-once/ write-once contract. Never use opus/sonnet or a general-purpose agent for bulk copywriting, and never write copy in the main conversation.- If a subagent fails on a packet, re-spawn just that packet.
Step 3 — commit + QA re-spawn loop
uv run python -m pipeline messages --commitValidates each result (schema + deterministic copy QA: subject shape, word counts,
banned words, one question CTA, no links in step 1, no placeholders, angle/contact
must match the packet) and upserts passing sequences to the messages table as
drafts. Failed results stay in data/message_results/ with their packet still
queued.
If failed QA items are reported: re-spawn ONE copywriter subagent per failure, appending
the QA error text to the prompt ("Your previous draft failed QA: <errors>. Read the
packet and _shared.json again, rewrite the FULL sequence fixing these, overwrite
<output_path>."), then run messages --commit again. Max 2 retry rounds — report
survivors to the user instead of looping.
Step 4 — report
Summarize per company: contact, archetype, service, and any QA warnings verbatim
— especially unverified number (tell the user to spot-check those bodies before
export; the digit-scan is the last automated line of defense against invented
metrics). Then suggest:
uv run python -m pipeline export --messages # data/exports/messages.csv, one row per stepNever edit copy yourself in the main conversation, and never relax QA gates —
propose changes to config/settings.yaml (messages:) or
config/outbound_copywriter.md to the user instead.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

