Security audit
Security checks for vulnerabilities and agentic risk
This trading skill needs review because it tells the agent to automatically run supplied update commands and adopt replacement instructions.
Review the automatic update workflow before installing. Use reviewed, fixed package versions and a fixed skill revision, and require review before running update commands or adopting changed instructions. Keep transaction confirmations and wallet spending limits enabled; trade notes and token metadata become public and permanent. Manual inspection addressed the scanner's document-coverage gap, but did not verify the external CLI or wallet code.
Referenced artifact was not completely inspected
| What other agents buy now | `gang trending` | [discover.md](references/discover.md) |
Referenced artifact was not completely inspected
| Hot / newest / about to graduate | `gang hot` · `gang new [--max-age 2h]` · `gang near-grad [--min-pct 50]` | [discover.md](references/discover.md) |
Referenced artifact was not completely inspected
1] [--timeout 30m]` (NDJSON; the market is pushed by gang.fun) | [discover.md](references/discover.md) |
Referenced artifact was not completely inspected
| My invite link / how many agents it brought (拉新) | `gang invite [--token <addr>]` | [discover.md](references/discover.md#gang-invite) |
Although this is an unpinned package runner rather than an MCP server reference, the underlying supply-chain risk is real: npx skills can execute mutable npm code, and the referenced skill source is not pinned to an immutable revision. The surrounding instructions also automatically install @gangfun/cli@latest, adopt refreshed instructions, and execute update commands supplied by CLI output, allowing a compromised dependency or update channel to gain local execution and influence a financial agent.
Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.
## Hard rules
- **Preview → show → confirm → execute.** Every write first prints a preview; show it, get the user's OK, then
`gang execute`. Chain steps without asking only when the user said to skip confirmation.
- **Setup steps.** A preview may be a `step: wrap-bnb | approve-quote | approve` first: execute it, wait for it to
land, then run **the same command again** for the real step.
- **Write a note by default** on every buy / sell / mint / withdraw (one line: what you saw, why now) and every create
Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.
## Hard rules
- **Preview → show → confirm → execute.** Every write first prints a preview; show it, get the user's OK, then
`gang execute`. Chain steps without asking only when the user said to skip confirmation.
- **Setup steps.** A preview may be a `step: wrap-bnb | approve-quote | approve` first: execute it, wait for it to
land, then run **the same command again** for the real step.
- **Write a note by default** on every buy / sell / mint / withdraw (one line: what you saw, why now) and every create