Config-free env mode, model recommendations, and provider docs by DavertMik · Pull Request #102 · testomatio/explorbot

DavertMik · 2026-07-19T20:50:02Z

What

Run Explorbot without an explorbot.config.js and keep the recommended models in one place.

Config-free environment mode

Set EXPLORBOT_* variables and Explorbot builds a config from the environment when no config file is found (a config file always wins). For CI one-liners, demos, and coding agents.

EXPLORBOT_AI_PROVIDER — a provider name; fills every role from its recommended models.
EXPLORBOT_AI_MODEL — pins the main model (a model id for the provider, or a standalone provider/model-id).
EXPLORBOT_VISION_MODEL / EXPLORBOT_AGENTIC_MODEL override those roles; EXPLORBOT_URL, EXPLORBOT_OUTPUT, EXPLORBOT_KNOWLEDGE[_FILE], EXPLORBOT_API_SPEC round it out.

EXPLORBOT_URL=https://app.example.com \
EXPLORBOT_AI_PROVIDER=openrouter \
  npx explorbot explore /login --max-tests 3

Config-free runs write to a temp dir, disable experience writing (experience.disabled), and turn the Historian off, so nothing is left in the project.

Model recommendations as a single source

models.json holds the recommended model / visionModel / agenticModel per provider. It is read at runtime (a bare provider name resolves through it) and ships with the npm package.

Generated provider docs

Each provider's export default config block in docs/basics/providers.md is generated from models.json by bunosh docs:models and verified in CI on release (--check). A > [!NOTE] is emitted when a provider is missing a role.

Mistral

Added as a supported provider — mistral-small-latest for model/visionModel (reads screenshots), mistral-large-latest for agenticModel.

Notes

Scoped to this feature only; unrelated in-progress work in the tree (demo-video skill, doc-collector, etc.) is intentionally left out.
bun test tests/unit/ — 836 pass.

🤖 Generated with Claude Code

Run Explorbot without an explorbot.config.js via EXPLORBOT_* variables. EXPLORBOT_AI_PROVIDER selects a provider and fills every role from its recommended models; EXPLORBOT_AI_MODEL pins the main model (a model id for that provider, or a standalone provider/model-id). Recommended models live in models.json, read at runtime (a bare provider name resolves through it) and shipped with the npm package. Each provider's config block in docs/basics/providers.md is generated from models.json by `bunosh docs:models`, verified in CI on release; a NOTE is emitted when a provider lacks a role. Adds Mistral as a provider. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

github-actions · 2026-07-19T20:59:38Z

Explorbot Self-Regression

Commit 8db4a5b · run

Scenario	Result	Attempts	Duration
basic (native)	PASS	1/3	7m
experience: control	OK — failed as expected	1/1	2m
experience: seeded	PASS	1/3	1m

Attempt details

basic (native) attempt 1 — PASS: login evidence: PASS (post-login plan=true, post-login research=true); research: PASS (files=6, wellFormed=true, keywords=7/3); scenarios: PASS (tests=8/5, features=4/3); tests passed: PASS (5 passed, 0 failed (reporter: 5 passed, 0 failed))
experience: control attempt 1 — PASS: control: OK — failed as expected (0 passed, 1 failed)
experience: seeded attempt 1 — PASS: seeded: PASS (1 passed, 0 failed)

Session analysis — basic (native):

Session Analysis

The Issues feature supports CRUD operations (create, read), search, and filtering by status and label. Core flows work: creating issues, searching by keyword, filtering by status/labels, and clearing filters all function correctly. The main friction point is an inconsistent form selector that caused retries during issue creation.

Coverage

Pages: /issues
Features: Issue list display, issue creation via toolbar, keyword search, status filtering, label filtering, filter clearing

What works

Issue creation — ET-1 Create new issue via toolbar and verify it appears in list
Keyword search — ET-2 Search for existing issue by title keyword and verify correct results
Status filtering — ET-3 Filter issues by status and verify results update correctly
Label filtering — ET-4 Filter issues by label and verify correct results
Filter clearing — ET-5 Clear search filters and verify all issues are displayed

Execution Issues

ET-1 Create new issue — first form selector .modal failed, retry with dialog#new-issue-drawer succeeded
ET-3 Filter by status — first "Select Closed status" click failed, passed on second attempt
ET-5 Clear filters — "All labels" click failed twice, but verification passed anyway (filter was already cleared)

…d docs EXPLORBOT_ENV_VARS in src/config.ts is the one place every environment variable is declared. The CLI renders it into help text on the root command and every subcommand (both boats included), so an agent running any `--help` discovers the full set. `bunosh docs:env` generates the tables in agentic-usage.md and commands.md from the same list, verified in CI on release. Adds EXPLORBOT_NO_BANNER, which was undocumented. Two tests guard the registry against drift in either direction. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

docs:models and docs:env become one docs:sync command with two non-exported helpers, so CI runs a single verification step. Signals staleness by throwing instead of calling process.exit, matching how regression:all already reports failure. A failed task() cannot be used for this: bunosh's test-environment guard checks `typeof Bun?.jest !== 'undefined'`, which is always true under Bun, so task failures never set exit code 1 here. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

One task per generated file, so a stale doc is reported by name. stopOnFail aborts the run at the first failure instead of accumulating a stale list and branching on it afterwards. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

DenysKuchma · 2026-07-20T09:43:24Z


 function buildExplorBotOptions(from: string | undefined, options: CLIOptions): ExplorBotOptions {
-  const sessionFile = options.session === true ? path.join(path.resolve(options.path || process.cwd()), 'output', 'session.json') : options.session;
+  const sessionDir = process.env.EXPLORBOT_OUTPUT ? path.resolve(process.env.EXPLORBOT_OUTPUT) : path.join(path.resolve(options.path || process.cwd()), 'output');


EXPLORBOT_OUTPUT is applied before determining whether config-free mode is active, so it changes the default --session path even when a config file is present
Need scope this override to env-config mode or derive the session path from the resolved config

DenysKuchma · 2026-07-20T09:45:22Z

+- **Experience is not written.** Nothing accumulates between runs, so a run is reproducible. Reading existing experience still works if the directory has any.
+- **The Historian is off.** No generated CodeceptJS or Playwright test files. Plans and reports are still written.
+
+For a long-lived agent that should learn across runs, point `EXPLORBOT_OUTPUT` at a stable directory or switch to a real config file.


env mode always sets experience.disabled: true
Remove this suggestion or enable experience writing when persistent output is explicitly requested.

DavertMik temporarily deployed to regression July 19, 2026 20:50 — with GitHub Actions Inactive

DavertMik had a problem deploying to regression July 19, 2026 21:27 — with GitHub Actions Error

DavertMik temporarily deployed to regression July 19, 2026 21:45 — with GitHub Actions Inactive

DavertMik had a problem deploying to regression July 19, 2026 22:25 — with GitHub Actions Failure

DenysKuchma requested changes Jul 20, 2026

View reviewed changes

Fix

cec876e

DenysKuchma temporarily deployed to regression July 22, 2026 09:31 — with GitHub Actions Inactive

DenysKuchma self-requested a review July 22, 2026 09:41

DenysKuchma approved these changes Jul 22, 2026

View reviewed changes

DenysKuchma merged commit 9ae94b9 into main Jul 22, 2026
3 checks passed

DenysKuchma deleted the feature/provider-model-config-and-env branch July 22, 2026 09:46

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Config-free env mode, model recommendations, and provider docs#102

Config-free env mode, model recommendations, and provider docs#102
DenysKuchma merged 5 commits into
mainfrom
feature/provider-model-config-and-env

DavertMik commented Jul 19, 2026

Uh oh!

github-actions Bot commented Jul 19, 2026 •

edited

Loading

Uh oh!

DenysKuchma Jul 20, 2026

Uh oh!

DenysKuchma Jul 20, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

DavertMik commented Jul 19, 2026

What

Config-free environment mode

Model recommendations as a single source

Generated provider docs

Mistral

Notes

Uh oh!

github-actions Bot commented Jul 19, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Explorbot Self-Regression

Attempt details

Session Analysis

Coverage

What works

Execution Issues

Uh oh!

DenysKuchma Jul 20, 2026

Choose a reason for hiding this comment

Uh oh!

DenysKuchma Jul 20, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

github-actions Bot commented Jul 19, 2026 •

edited

Loading