High-risk tools in Devbox

Admin operations. action: invite_alpha | list_alpha_users | transition_alpha | stripe_platform_setup | stripe_platform_status | stripe_set_key | disputes | refunds.

Agent handoff: send summary + context to DevBox, get next action (continue, create_run, done, wait). Use when about to summarize for user — DevBox tells Cursor what to do next. ...

Billing and subscription management. action: subscription_status | usage | checkout | portal | entitlements | refunds | budget. Requires authentication.

Resolve a canonical bundle into provenance-heavy derived context. Returns ordered semantic sections, inline provenance, and compact supporting evidence so agents see compressed ...

Compile a context bundle from a set of artifacts, applying layering (base, project, session, ephemeral), deduplication, and budget trimming. Returns a ready-to-inject bundle wit...

Explain why a context bundle was compiled the way it was. Returns a human-readable explanation of layering decisions, deduplication, trimming, and trust-tier filtering applied t...

Deploy and rollback. action: deploy | rollback. deploy requires step_id, image_uri, digest, service_name, environment. Requires prior approval for gated steps.

Project Health Map. action: get | analyze | feature | summary | prompt | outcome | delta | infer_goals. get: latest analysis with all features. analyze: trigger new analysis. fe...

Get DevBox tool documentation for a topic. Topics: turn_start, workflow, deploy, runs, projects, approvals, ci, logs, planner, memory, project_state, handoff, plans, terminology...

Project memory for agents. action: search | store | bootstrap | bootstrap_status. search: semantic retrieval. store: add a memory. bootstrap: start phased scan (docs→config→code...

Ops and diagnostics. action: runner_health | worker_stats | ecs_status | logs_control_plane_tail | trace_by_run_id. Use trace_by_run_id for run-level trace inspection (more reli...

Project plans for multi-stage work. action: create_plan | get_plan | update_plan_stage | list_plans. Coordinates multiple DevBox runs under a single plan.

Project state and pipeline. action: state | pipeline_status. state returns run_branches, open_prs, orphans, pipeline_summary. pipeline_status returns deployment pipeline stages ...

Quality Graph queries — the unified verification chain from outcome to evidence. action: chain (full chain for a feature) | gaps (all gap nodes by severity) | failing (all faili...

Submit code result or instruction feedback. action: code_result_submit | instruction_feedback. code_result_submit: REQUIRED workspace_path when branch provided. instruction_feed...

Get full run guidance in one call: run status, next instruction, pending approvals, and suggested_action (APPLY_CODE_INSTRUCTION, DECIDE_APPROVAL, WAIT, DONE). Use this when the...

Run lifecycle control. action: cancel | close_with_note | report | diff | advance | next_instruction | events_poll. Use close_with_note when agent resolved issue outside normal ...

Link an existing PR to a run. Use when PR was opened from a different branch than the run's repo_ref. Syncs run with PR so the run page shows the PR link.

Run step logs. action: step_log_get | logs_tail | step_fail_orphan. step_fail_orphan marks stuck step as orphaned when control plane restarted.

Step control. action: retry | complete | rewind. retry: reset failed step to queued so worker re-executes. complete: manually mark step success (e.g. deploy done outside DevBox)...

DP-05: Run validations. action: review_code | review_architecture | uat_simulate | validate_logs | run_checks | check_coverage | generate_checks | verification_trends | verifica...

Run discovery and creation. action: list | create | get | plan | steps. list discovers runs with awaiting_delegation; create starts a new run. REQUIRED at turn start: list to fi...

Surface verification. action: surfaces_dashboard | surfaces_crawl | surfaces_verify | surfaces_fix | get_next_fix | list_issues. get_next_fix: returns highest-priority unresolve...