Repository navigation
Let users choose the model for each agent #45
Description
Problem: Only the api agent supports choosing a model (SECFOO_API_MODEL). The CLI adapters (claude, codex, gemini, copilot, cursor, droid, ...) build a fixed argv in build_command() and never pass a model, so a scan always uses the CLI's own default. The only workaround is changing that CLI's global config, which affects every other use of the CLI too.
Proposal:
- Optional per-agent setting in
~/.secfoo/config.toml, e.g.[agents.claude] model = "claude-sonnet-5-5"
- A
--modeloption onsecfoo runthat overrides the config value for one run. - Each adapter appends its CLI's own flag only when a model is set (
--modelfor claude/codex/gemini/copilot/cursor,-mfor droid; check each CLI's--help). Adapters whose CLI has no model flag should warn and ignore the setting. - Store the model on the run record so results from different models can be compared.
Where to look: src/secfoo/agents/base.py, the adapters in src/secfoo/agents/, src/secfoo/config.py, src/secfoo/cli.py (run).
Done when: unit tests in tests/test_agents/ cover the argv with and without a model for every adapter, and pytest tests passes.
Activity
karthickmk2001 commentedon Nov 11, 2026
I'd like to take this. Working on it now; a PR will follow.
karthickmk2001 commentedon Nov 11, 2026
Correction to my comment above: this was already done by @kushaldave002 in #60 (merged 2026-11-08). It adds
[agents.<name>].modelin config, a--modeloverride onsecfoo run, each adapter's own flag (--model/-m) with a warning for adapters that have none, and the model stored on the run record. Adapter argv tests are intests/test_agents/test_models.py. The PR didn't link this issue, so it stayed open. Closing as completed.docs: add a Closes # line to the PR template
docs: add a Closes # line to the PR template
docs: add a Closes # line to the PR template
leaderboard: update 2026-11-11T20:44Z (credit #60->#45, #61->#47)
docs: add a Closes # line to the PR template