Imported from tamaltm/signalscope (
AGENTS.md). Install upstream withnpx skills add tamaltm/signalscope. Copyright stays with the author.
AGENTS.md — SignalScope
How to fill this in: write only what an agent could NOT work out by reading the repo. Skip the directory tree (
lsshows it), the dependency list (the manifest shows it), and generic advice like "write clean code" or "handle errors" — a capable model already does those, and every line here is loaded into context on every single session. If you find yourself describing the code, delete it. If you find yourself describing something that once cost someone an afternoon, keep it.
Project
- What this is: An original mobile prototype for scanning, filtering, and understanding fictional insider-activity signals.
- Who it is for: Mobile users and ArkLab AI assignment reviewers
- Current phase: Phase 6 testing and refinement complete; Phase 7 Android packaging and submission artifacts next
Commands
npx eas-cli@latest build --platform android --profile preview— thepreviewprofile must produce an installable APK, not the default Play Store AAB.
Read first
docs/PRD-*.md(what we're building — the source of truth)docs/TechDesign-*.md(how we're building it)agent_docs/project_brief.mdagent_docs/tech_stack.mdagent_docs/testing.md
If this file or agent_docs/ still has unresolved template placeholders, fill them from
the two docs above before planning. Load anything else only when it becomes relevant.
Gotchas
- Every displayed company, person, ticker, value, date, signal, and chart point must be invented local demo data. Do not add financial APIs, copied source content, or real filings.
- Keep exactly three application screens: Home, Screener, and Trade Details.
- Purchase/Sale meaning must use text and direction icons as well as color.
- The required disclaimer in the PRD must be reproduced exactly.
Protected areas — ask before changing
.env*, secrets, credentials, private logs.github/workflows/, deployment, infrastructure- existing database migrations
- auth, payments, billing, production email/send flows
- AI provider credentials, MCP servers, tool permissions
Never print, commit, or transmit secrets, tokens, private logs, or production data. Never delete files, rewrite large areas, or change infrastructure/auth/billing/migrations without approval.
Done means
Report: files changed · commands run · test/build/device results · AI eval evidence if applicable · remaining risks · rollback notes if relevant.
When this file gets long, that is the signal to split it. Move task-specific
procedures (deploy steps, release checklists, API references) into
.claude/skills/<name>/SKILL.md, where only the one-line description stays in
context and the body loads when it is actually needed. Move
directory-specific conventions into <subdir>/CLAUDE.md, which loads only when
work touches that directory. Keep universal constraints and safety prohibitions
here — never move a "never do X" rule somewhere it might not be loaded.