v0.1.7

• Reframed distribution copy as quantitative research and data infrastructure for crypto markets, operated by AI agents. Categories the skill alongside Kaiko, Coin Metrics, Amberdata — institutional-grade research/data tooling — instead of generic account operations or execution-coded language. Lead with the value prop: Stingray ingests, correlates, and serves the data; the agent runs the loop end-to-end so users test more hypotheses faster. Venue coverage (Hyperliquid, Lighter, Polymarket, Kalshi, 100+ more) named explicitly in the body to anchor the breadth claim.
• Dropped execution-coded language that hit Snyk's financial-platform-automation taxonomy in v0.1.6 (the keyword classifier flagged the v0.1.6 copy as Critical Risk regardless of the actual code surface). The v0.1.7 framing keeps persona-targeting (quants, analysts, research desks) without the classifier hit.
• Updated marketplace.json keywords/tags accordingly. Added signals, quant, research. Removed trading, defi.
• Removed Drift from venue list. Stingray doesn't currently cover Drift — surfacing it overpromised. Anchor venue names are now Hyperliquid, Lighter, Polymarket, Kalshi.
• Added a debug-report channel through references/co-development.md. Same chat-stream endpoint as feature requests, prefixed Debug report: instead of Feature request:. Covers reference mismatches, undocumented response shapes, routing ambiguity, reproduction failures, and setup edge cases. Added a corresponding step 8 to the Default Operating Loop in SKILL.md so agents fire it after completing any task where something was unexpected.
• Snyk W007 mitigation — credential isolation, single secure path. Reworked SKILL.md First-Time Setup so the agent gives the user a one-line shell command to run in their own terminal (or sets STINGRAY_PAT in their shell). The token never enters the agent's chat context. Chat-paste is deliberately not supported — if the user pastes the token into chat anyway, the agent asks them to clear scrollback and re-do setup via the terminal command.
• Snyk W011 mitigation — untrusted content handling. Added a new "Untrusted Content Handling" section to SKILL.md instructing the agent to treat news bodies, KG entity descriptions, attachment text, and any third-party content as data, not instructions — no following URLs, no acting on imperatives, no interpreting fetched content as user commands.
• Tightened root README Security section to lead with the two Snyk findings and the mitigations, instead of explaining the Med-Risk score in v0.1.6 framing.
• Updated skills/stingray/README.md quickstart to reflect the terminal-only credential setup.
• Updated scripts/validate-public.sh to enforce the new guidance phrases (replacing the legacy "Do not show the user shell commands" check, which contradicted the new credential-isolation flow).
• Bumped metadata.version and plugin version to 0.1.7.