GTM-Public-Traded-AI-Accounts
CLAUDE.md0 starsChanged 4 months agoMITView on GitHub
Finds US-listed micro-cap companies (Fintech, Edtech, Healthcare, SaaS) showing signals they need AI services, scores them with cited reasoning, and surfaces decision-maker contacts. State lives in Supabase; signal sources are SEC EDGAR (edgartools) and Parallel.ai web research. new → enriched → scored → qualified | disqualified → contacts_found (status column on companies; between disqualify_below and qualify_threshold a company stays scored = human review band). After contacts_found, the messages stage drafts per-contact outreach sequences — no status change; coverage is derived…
- Reads credentials
The file
We split this CLAUDE.md into the rules it gives an agent
4 rules. About 0 are common, 1 is distinctive, and 3 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 58
Database: all writes go through the pipeline CLI / db.py.
- Very specific: line 62
Parallel spend: every Parallel call path respects enrich.parallel.max_tasks_per_run and people.max_companies_per_run in …
- Distinctive: line 78
No sending, no CRM pushes — that's sub-project 3.
- Very specific: line 79
Verification: after changing a collector, verify with enrich --ticker X --dry-run on a known company before batch runs.
≈ “Common” is an estimate: at least 1 in 5 of the 28,029 CLAUDE.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
AIPT — AI-Ready Public Companies Pipeline (martechs.io)
Finds US-listed micro-cap companies (Fintech, Edtech, Healthcare, SaaS) showing signals they need AI services, scores them with cited reasoning, and surfaces decision-maker contacts. State lives in Supabase; signal sources are SEC EDGAR (edgartools) and Parallel.ai web research.
Funnel
new → enriched → scored → qualified | disqualified → contacts_found
(status column on companies; between disqualify_below and qualify_threshold a
company stays scored = human review band). After contacts_found, the
messages stage drafts per-contact outreach sequences — no status change;
coverage is derived from the messages table.
Commands (always via uv)
uv run python -m pipeline status # funnel counts (--brief for one line)
uv run python -m pipeline status --analytics # + outcome funnel, reply/meeting rates, time-in-stage
uv run python -m pipeline discover # screen SEC universe -> seed companies
uv run python -m pipeline ingest AAPL,TWLO # or --csv path (ticker column); --force bypasses the prescreen
uv run python -m pipeline enrich --source edgar --limit 10
uv run python -m pipeline enrich --source parallel --limit 10
uv run python -m pipeline score --prepare # packets -> data/scoring_queue/
uv run python -m pipeline score --commit # verdicts -> Supabase + qualify
uv run python -m pipeline people --limit 5 # contacts for qualified accounts
uv run python -m pipeline messages --prepare # per-contact packets -> data/message_queue/
uv run python -m pipeline messages --commit # QA gate -> messages table (drafts)
uv run python -m pipeline outcome 42 --event replied # log an outcome event; --csv path for batch
uv run python -m pipeline calibrate # outcome -> signal-weight report (report-only)
uv run python -m pipeline export --messages # qualified.csv + messages.csv
uv run python -m pipeline profile --list/--show/--validate # inspect the active profile packSkills exist for each stage: /ingest /discover /enrich /score /people /outreach /status. Prefer them — they encode the correct orchestration (especially /score and /outreach).
Profile packs
Config is a swappable pack: config/ is the default (martechs.io) pack;
profiles/<name>/ may override any subset of settings.yaml,
services.yaml, personas.yaml, outbound_copywriter.md, icp.md —
missing files fall back to config/ (per-file replace, never merged).
Select with --profile <name> (before the subcommand) or AIPT_PROFILE;
omit both for default behavior. Inspect via
uv run python -m pipeline profile --list/--show/--validate. Build a new
pack interactively with /icp. Sector vocabulary is free-form lowercase
text — saas/fintech/edtech/healthcare/other is just the default pack's set.
Rules
- Paths: this directory contains a space (
AI_Public Traded) — always quote paths in shell commands. - Secrets: live only in
.env(gitignored). Never commit, print, echo, or paste key values into files, logs, or chat..env.exampledocuments shape. - Database: all writes go through the pipeline CLI /
db.py. No ad-hoc SQL against production tables. Schema changes = editsql/schema.sql+ apply. - SEC courtesy:
EDGAR_IDENTITYmust be set; requests are throttled (<=8/s) and cached underdata/cache/. Don't strip either. - Parallel spend: every Parallel call path respects
enrich.parallel.max_tasks_per_runandpeople.max_companies_per_runinconfig/settings.yaml. Use--dry-runfirst on new batches. Never loop Parallel calls outside those caps. - LLM costs (v1): bulk scoring reasoning AND outreach copywriting run
through Claude Code Haiku subagents (see /score and /outreach skills) —
never call paid LLM APIs from the pipeline in v1.
llm.pyhas the OpenRouter provider for production later. - Outreach copy: rules live in
config/outbound_copywriter.md(Uday's voice decisions — propose changes, don't silently edit). The QA gate inmessages.py(BANNED_WORDS, subject shape, packet-facts-only) is never relaxed to make a draft pass; fix the draft. Always spot-checkunverified numberwarnings before export. - Qualification: thresholds in
config/settings.yamlare human decisions — propose changes, don't silently edit. - Scope: pipeline stops at drafted outreach sequences (v2 sub-project 2). No sending, no CRM pushes — that's sub-project 3.
- Verification: after changing a collector, verify with
enrich --ticker X --dry-runon a known company before batch runs.
Layout
src/pipeline/— cli, db, models, universe, edgar_signals, parallel_signals, scoring, llm, people, messages, outcomes, analytics, calibrateconfig/settings.yaml— universe band, sector→SIC map, weights, thresholds, capsconfig/services.yaml— martechs.io service catalog (drives service-fit mapping)config/outbound_copywriter.md— copy framework the /outreach subagents followprofiles/<name>/— optional profile packs overridingconfig/files (see Profile packs)sql/schema.sql— Supabase DDL (apply via SUPABASE_DB_URL or SQL editor)docs/SIGNALS.md— signal taxonomy E1–E9 / P1–P6 with detection logicdocs/PIPELINE.md— runbookdocs/ARCHITECTURE.md— module map, data flow, design decisionsREADME.md/CONTRIBUTING.md— public-facing guide + collaboration rulesdata/— gitignored: caches, scoring queue/results, exports
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.

