praxist-takeover

name

praxist-takeover

description

Orchestrate first-use Praxist onboarding, current task initialization, validation, and detached run launch from a supported agent interface. Use when a user wants a one-command or one-conversation path from an existing runnable research project to a started Praxist run, asks to onboard plus initialize plus start, asks for repo-to-task-to-run setup, or wants the agent to prepare and launch Praxist with minimal interaction. This skill composes onboarding, full task initialization or task-harness repair, runtime checks, control start, and confirmation gates without changing Praxist core.

Praxist Takeover

Use this skill to move a user from an existing runnable research project to a detached Praxist run in one guided flow. It composes existing skills rather than duplicating their detailed procedures:

• praxist-onboarding for Praxist context and local readiness scan.
• praxist-task-initialization or praxist-interactive-task-init for task-project creation.
• praxist-control for task validation and detached launch.

The operator-facing praxist --takeover command is only a first-project handoff into this skill. It does not replace, abbreviate, or independently implement any step below. Treat its selected absolute project path and final Enter confirmation as explicit operator context.

The first operational action must be the complete praxist-onboarding workflow. Do not replace onboarding with a shorter local checklist, and do not proceed to task initialization or launch until onboarding has produced its readiness summary. This applies even when Codex already appears familiar with Praxist from prior context.

When no launch-ready task exists, the next action must be the complete current praxist-task-initialization workflow. praxist-takeover must invoke and follow that skill rather than recreating a shortened initializer from this file. Load and read the installed skill's current SKILL.md at execution time; do not rely on a remembered or summarized task-init procedure. Wait for task-init's evaluator, lane-routing regression, resource plan, and required validations to finish before continuing. A user who invokes only praxist-takeover must receive the same harness contract as a direct task-init user.

Hard Preconditions

Print this before doing work:

**IMPORTANT PRECONDITION**
This takeover assumes the research project already runs on this machine,
with required code, data, simulator, runtime/container, and credentials present.
If those are missing, initialization must stop instead of creating a weak Praxist task.

Workflow

1. Run the complete onboarding flow:

   • invoke praxist-onboarding and follow its SKILL.md procedure;
   • let onboarding inspect nearby Praxist source or pip installs, praxist CLI availability, installed skills, provider-key presence by name only, and bounded local environment context;
   • keep onboarding read-only and bounded as defined by that skill;
   • stop if onboarding reports that Praxist itself is unavailable or unusable.

2. Resolve the research project path:

   • use the user-provided path if present;
   • otherwise use the current working directory.
   • if that path is a bundled example inside a Praxist source checkout or package resource, do not initialize or launch it there. Identify its name with praxist examples list, run praxist examples install <name>, report the writable destination, and continue from that destination. Preserve an existing destination unchanged.

3. Decide whether a runnable task directory already exists:

   • if an explicit task directory is in context, validate that path;
   • otherwise scan nearby roots up to depth 3 for task.yaml, excluding .git, .venv, node_modules, experiments, run directories, caches, and raw data roots.

4. If no valid task exists, initialize one:

   • load the currently installed praxist-task-initialization skill and complete it for the default low-friction path; do not substitute a reduced praxist-takeover implementation or an older remembered checklist;
   • use praxist-interactive-task-init when the user asks for confirmation, an interactive flow, strict goal confirmation, or when the task objective/constraints are ambiguous. The interactive path must still delegate task construction and final validation to the current task-init workflow.

5. Resolve and show the absolute task path before launch; confirm the exact task path when selection is ambiguous or the user has not authorized launch. If the user's current instruction already names or unambiguously identifies the task and explicitly authorizes direct/no-confirm launch, obey it without adding another confirmation round.

6. Check and warn about durable evidence-retention lanes before launch:

   • read the protocol-intent decisions produced by current task-init. User instructions own which evaluator stages may launch, rank, count as mature, become parents, or satisfy close. Do not impose a full-protocol-only rule; an explicitly authorized reduced/partial/scout protocol is valid when its actual stage and effort/coverage remain visible and every downstream policy agrees. Only undeclared drift must be prevented from impersonating mature evidence;

Refusal Conditions

Stop instead of launching when:

• no runnable research project can be identified;
• required local code, dataset, simulator, or runtime is missing;
• no valid task directory exists and task initialization aborts;
• the selected task path is ambiguous or not confirmed by the user;
• praxist resolve fails;
• the selected task has an unreachable configured parent-eligible lane or its lane-routing regression fails, regardless of whether the task is new or pre-existing;
• the current user or task explicitly requires an accelerator but the task interpreter, public evaluator canary, and scheduler profile cannot establish one coherent execution backend;
• required provider/runtime configuration is absent and no fallback is declared.

Output

End with a compact launch summary:

Item	Value
Research project	absolute path
Task path	absolute path
Praxist command	redacted command
Run id	id or unavailable
Run dir	absolute path
PID/state	value from start/status
Provider/runtime	selected values
Lane routing	task-local regression result and reachable parent lanes
Monitor	praxist --monitor --run-id <run_id>; foreground TUI is not opened automatically
Next command	praxist-control with a research-status request; praxist --monitor --run-id <run_id>

Do not edit Praxist core, guard code, provider credentials, or existing run artifacts in this skill. Do not run long baseline measurements unless task initialization explicitly asks the user and receives approval.