#363
vercel-cli-with-tokens: token checks print the token into the agent transcript
At 1a2c79d, several steps in skills/vercel-cli-with-tokens/SKILL.md print the Vercel token value into the agent's output, or ask the user to paste it into chat. Agent transcripts are often logged, shared or synced, so the token ends up stored outside the user's shell.
- Line 20,
printenv VERCEL_TOKEN, prints the token to check whether it is set. - Line 28,
grep '^VERCEL_TOKEN=' .env, prints the full line, value included. - Lines 42 and 78 (
grep -i 'vercel' .env) and line 317 (printenv | grep -i vercelandgrep -i vercel .env) print every matching variable with its value, which includes any token. - Line 53 says: "ask the user to provide one". In an agent session the user then pastes the token into the chat.
The skill already keeps the token out of --token arguments (lines 55-67). These checks are the remaining way it leaks.
Suggested changes:
- Test for presence without printing the value, for example
[ -n "$VERCEL_TOKEN" ] && echo "VERCEL_TOKEN is set". - For
.envfiles, list variable names only, for examplegrep -i vercel .env | cut -d= -f1. Export the value without echoing it, as line 34 already does. - Replace line 53 with an instruction to the user: set
VERCEL_TOKENin their own shell or.envfile, and do not paste it into the conversation.
Proposed patch for this issue. It is the complete change and can be applied with
git apply.