robot_sf_ll7

Open a conservative Robot SF PR with scope verification, freshness checks, and artifact discipline.

Use for an autonomous Robot SF issue-to-PR loop that selects eligible GitHub issues, implements one scoped issue at a time, validates, pushes, and opens PRs.

Guarded PR merger; merges merge-ready PRs after verifying label, CI status, branch protection, and preflight checks.

Use for an autonomous Robot SF PR review loop that fixes scoped review gaps, validates proof, resolves review threads, and applies merge-ready; not for merging.

Autonomous issue-to-PR workflow from next eligible issue to ready PR with consistent metadata handling.

Continuous goal autopilot; orchestrates implement, review, merge, and discover cycles with preflight validation and delegation failure recovery.

Clarify ambiguous GitHub issues by tightening scope and acceptance criteria, proposing solution options with pros/cons, and marking decision-required issues when maintainer input is needed.

LLM-backed review workflow for Project #5 priority inputs; assess plausibility, propose values with uncertainty, route maintainer-value tradeoffs to issue-audit, and optionally apply explicit opt-in updates.

Review existing GitHub issues against the repo's issue-template contract and repair underspecified issues when the fix is clear.

Review merged PRs after the fact to decide whether autonomous routing produced useful progress, partial coverage, duplicate coverage, or a successor slice.

Create structured GitHub issues from vague prompts using repo templates, conservative assumptions, and Project #5 metadata.

Submit generic SLURM campaigns with preflight, config provenance, job metadata, artifact expectations, and failure classification.

Keep one skill-owned Robot SF learning or training SLURM job active by selecting the best current experiment candidate, closing implementation gaps through an issue-to-PR workflow, and submitting the validated job from its owning worktree.

Fix GitHub PR review comments with branch-safe edits, validation, and explicit thread resolution.

Run the repository PR readiness pipeline using shared scripts/dev entry points (ruff fix/format, parallel tests, coverage, and docstring checks).

Clean up the current branch in the Robot SF repo by following docs/dev_guide.md and reusable scripts/dev commands; use when asked to tidy a branch, run Ruff format/fix, or run parallel pytest before sharing changes.

Analyze a camera-ready benchmark campaign for consistency, runtime hotspots, fallback/degraded planners, and reproducibility metadata.

Synthesize multiple issues, configs, seeds, metrics, and artifacts into conservative mechanism-level conclusions with caveats.

Draft or review benchmark and manuscript-support docs conservatively, with explicit provenance, reproducibility, and caveat handling.

User-in-the-loop open-issue audit that asks one readiness-blocking question at a time or one priority-tradeoff question at a time and updates issues as decisions are made.

Maintain GitHub issue contracts through template audits, ambiguity clarification, and user-decision application.

Submit issue-791-specific Auxme training jobs with explicit config provenance and wrapper-safety checks.

Submit issue-791 style Auxme SLURM jobs with explicit config, live partition pressure checks, and max-time-safe wrapper routing.

Split a parent, epic, decision, or research issue into the smallest independently implementable child issue with duplicate checks and conservative parent linking.

Review exact planner/scenario/seed/episode traces and videos without overgeneralizing from qualitative samples.

Generate a focused repository context map before multi-file changes; use when you need to identify the relevant files, docs, commands, and risks.

Choose the most appropriate repo-local skill for an ambiguous task by consulting .agents/skills/README.md.

Use for an autonomous Robot SF issue-discovery loop that finds bounded improvement opportunities and creates evidence-graded GitHub issues; not for implementation.

Capture private candidate lessons from agent execution into `.git/codex-agent-runs/notes/inbox/` when a repeatable workflow, routing, validation, tooling, or instruction improvement is noticed.

Promote accumulated private `.git/codex-agent-runs/notes/inbox/` workflow lessons into small, evidence-backed repository instruction, skill, docs, or tooling changes with validation.

Maintain a clear next-work queue in GitHub Project #5 by normalizing issue status, priority, and execution order; route genuine priority tradeoffs to issue-audit.

Repo-wide risk-proportional validation workflow for non-trivial changes with context, risk, validation, and follow-through.

Verify branch changes against origin/main with claim-based evidence, not only test status.

Analyze adversarial route/search campaigns with canonical commands, expected artifacts, status boundaries, and claim limits.

Evaluate and improve AI workflow outputs with small goldens, rubrics, and repeatable checks; use when tuning skills, prompts, instructions, or agent behavior.

Analyze latest policy analysis sweep runs (*_policy_analysis_*) by comparing episodes/summary metrics, diagnostics, and video artifacts; generate a concise markdown report and optional frame snapshots.

Classify, promote, or document generated artifacts so durable evidence is separated from local output caches.

Focused measurement-aware refinement loop for Robot SF prompts, docs, and small code changes; use when a task benefits from trying a few simple improvements.

Autonomous iterative experimentation loop for measurable Robot SF tasks; use when the user wants an improvement loop with baseline, experiments, and keep/discard decisions.

Fast benchmark-faithful orientation for scenario splits, baselines, metrics, artifacts, and reproducibility constraints in robot_sf_ll7.