docs + run: phase agents look for production credentials wherever the repo's docs name them; document it, and offer a deny list
Maintainers usually reply within 1 day
Nobody has claimed this yet.
Assessment
- Difficulty
- 3/5
- Estimated time
- 1-2 days
- Newbie friendliness
- 35/100
- Issue type
- Documentation
- Clarity
- Mostly clear
- Activity status
- Active
- Tech stack
- nodejs, supabase, typescript
- Domain
- devtools, documentation, security
Research direction
Start with the README security section referenced in the issue and the phase prompts that name .env.local and live Evidence commands in the exec and qa skills. Read how run.denyPaths would map to the driver's sandbox configuration before touching prompts. Done means the README states the permissions-skipped behavior and a deny-path setting exists with a default-empty test.
Written by the indexing model from the issue text.
Description
What happens
On a repo whose scripts and docs name .env.local, vercel env and a Supabase project, sequant's spec, exec and qa agents (running, as sequant runs them, with permissions skipped) try to use those credentials to do the work the issue describes: run the audit script against the live database, check production's environment variables, verify a change on the live site. They look in the worktree, in the main checkout, in the parent directory, in the user's home, and in the Vercel CLI's token store. Nothing in sequant stops them; in our replay a sandbox-exec profile and the absence of the files did.
Evidence (sequant 2.20.0 at 31c6f8bd, 15 F runs on a private mirror of a repo that reaches production, 2026-11-02 to 2026-11-03)
A scan of every Bash call in the phase transcripts, with the call's result:
ls .env.local,[ -f .env.local ],ls -a | grep -c "^.env.local$": existence checks in most runs; the file was absent.grep -c "^SUPABASE_SERVICE_ROLE_KEY=" .env.local: a read attempt (#1143 F5); no file.npx tsx --env-file=.env.local scripts/dev/query.ts "sb.from('shops').select(...)"in a loop of six live queries (#1144 F4), andnpx tsx --env-file=/Users/…/matcha-maps-e2c/.env.local scripts/dev/query.ts …(#1144 F1): node refused at start,.env.local: not found, exit 9.vercel env ls production(#1153 F1) andvercel env ls(#1153 F5): the sandbox refused the CLI's token directory,EPERM: operation not permitted, mkdir '~/Library/Application Support/com.vercel.cli'. Outside the sandbox the CLI is logged in; the call would have listed production's variables.ls /Users/…/matcha-maps/.env.local(the real checkout, not the worktree) andtest -f ~/.claude/secrets.env && grep -cE '^(export )?(NEXT_PUBLIC_SUPABASE_URL|SUPABASE_SERVICE_ROLE_KEY)=' ~/.claude/secrets.env(#1153 F2): bothOperation not permittedunder the sandbox.scripts/fix/null-places-photos.ts --verify-live --sample 6andcurl -s https://getmatchamaps.com/: anonymous GETs of the production site, in 14 commands across the batch.
The native comparison arm on the same issues (a claude -p build session plus /code-review --fix, also with permissions skipped, given the same issue and the same Evidence lines) ran an ls .env.local-style existence check in 4 of 15 runs and nothing else on this list: no --env-file command, no live query, no Vercel call, nothing outside its worktree, no fetch of the site. The difference is the phase prompts: sequant's qa asks for each AC's Evidence to be run, and the Evidence lines name --env-file=.env.local and live checks.
Ask
- In the README's security section (#1028): say that phase agents run with permissions skipped and will use any credential the repo's own scripts name, including files outside the worktree and CLIs logged in on the host, and that a replay, a benchmark or a borrowed machine needs an OS-level sandbox.
- A
run.denyPathssetting (or--deny-path, repeatable) that sequant turns into the driver's sandbox configuration where the driver has one, and that the qa and exec prompts receive as "do not read" so the agent does not spend turns trying. Default empty; the docs give~/.claude/secrets.envand the main checkout's.env.localas the first entries. - Smaller: the exec and qa skills could treat an Evidence command that names
--env-fileor avercel/supabaseCLI as "needs a human", the way qa already treats CI (#1630), instead of attempting it.
Why it matters
Sequant's pitch includes unattended runs. On a repo with production reach, an unattended run today is one that will try the production credentials if the issue's Evidence tells it to. In our replay every attempt was stopped by things sequant does not provide.
From a private replay study of sequant on our own repos; raw runs are not public. Filed by the owner.
- Dominant language
- TypeScript
- Stars
- 1
- Forks
- 1
- Avg merge
- 11h 3m
- Merged PRs (30d)
- 84
Getting set up
- No Dockerfile or Docker Compose file
- No pull request template
- Read the contributing guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
More from sequant-io/sequant
-
graph-2026-11-simplify
Difficulty 2/5 1-3 hours Newbie friendliness 72/100
sequant-io/sequant#1650 ·
Maintainers usually reply within 1 day
-
run: --phases does not suppress the UI-label test phase (determinePhasesForIssue adds it to an explicit list)Possibly taken @admarble claimed this today. Open
Difficulty 2/5 1-3 hours Newbie friendliness 72/100
sequant-io/sequant#1631 ·
Maintainers usually reply within 1 day
-
Difficulty 2/5 1-3 hours Newbie friendliness 75/100
sequant-io/sequant#1601 ·
Maintainers usually reply within 1 day
-
bug
Difficulty 2/5 1-3 hours Newbie friendliness 86/100
sequant-io/sequant#1582 ·
Maintainers usually reply within 1 day
-
Difficulty 1/5 Under an hour Newbie friendliness 76/100
sequant-io/sequant#1381 ·
Maintainers usually reply within 1 day
All issues in sequant-io/sequant
Similar issues
-
content triage
Difficulty 2/5 1-3 hours Newbie friendliness 72/100
Maintainers usually reply within 1 day
-
Difficulty 2/5 1-3 hours Newbie friendliness 82/100
tensorflow/tfjs#8754 ·
-
tool-submission
Difficulty 1/5 Under an hour Newbie friendliness 83/100
BraveOPotato/NoSignups#2348 ·
-
Difficulty 2/5 1-3 hours Newbie friendliness 78/100
code-yeongyu/oh-my-openagent#12784 ·
Maintainers usually reply within 1 day
-
external-issue to-triage
Difficulty 2/5 1-3 hours Newbie friendliness 71/100
Maintainers usually reply within 1 day