f0ea84fbd4
fix(runner): recursively scan repository root to locate nested .vibncode/tasks directory
2026-06-03 15:00:27 -07:00
4f9c82b525
fix(runner): support optional leading dashes in markdown checkboxes
2026-06-03 14:56:40 -07:00
71bea9103f
fix(runner): parse tasks from app-specific workspaceRoot instead of repoRoot
2026-06-03 14:52:22 -07:00
76c161eedf
fix(runner): support recursive package.json scanning in build verification
2026-06-03 14:28:13 -07:00
fa0e460c1c
fix(runner): raise SUB_MAX_TURNS to 40 and inject Surgical Healing Protocol into subtasks
2026-06-03 14:15:37 -07:00
fa8a919214
fix(runner): fix fingerprint collision on projectId and relax loop-break limits
2026-06-03 13:58:56 -07:00
ba2eaa55f2
feat(runner): implement state-of-the-art task-by-task meta-loop for offline delegation
2026-06-03 13:46:49 -07:00
ae54954545
feat(runner): increase MAX_TURNS limit to 80 for large task sessions
2026-06-03 13:13:50 -07:00
d16ef9c138
feat(runner): implement Surgical Healing Protocol in Ralph Loop
2026-06-03 13:03:10 -07:00
3d07cf38b6
fix(runner): wire ToolContext vibnApiUrl + mcpToken so agent tools reach the frontend MCP
...
buildContext() hardcoded vibnApiUrl='http://localhost:3000 ' and mcpToken='',
so every agent tool call (projects_list, workspace_describe, apps_list, ...)
fetched the runner itself on a dead port and failed with 'fetch failed'.
Now /agent/execute reads mcpToken from the request body and sets
ctx.vibnApiUrl (from VIBN_API_URL), ctx.mcpToken, and ctx.projectId before
running the agent.
2026-05-30 19:15:43 -07:00
2e5ca29c40
fix(runner): bump max tool rounds to 30 to support long deployments
2026-05-19 14:51:11 -07:00
8071ac9049
fix(runner): remove leftover syntax errors
2026-05-19 14:34:42 -07:00
2f86a4262e
fix(runner): resolve TypeScript compilation errors
2026-05-19 14:14:34 -07:00
abf9bf89c2
chore: convert submodules to standard directories for true monorepo structure
2026-05-13 14:54:23 -07:00
