linzhe001
34 verified skills34 total stars
release
Codex wrapper for WF10 release and submission packaging. Use when the user wants validation, packaging, or submission preparation according to the original workflow.
development1
code-expert
Codex wrapper for WF7 first-pass code generation. Use when the user wants implementation generated directly from `project_map.json`, the roadmap, and the original Claude skill contract.
development1
code-debug
Codex wrapper for post-WF7 code modification and debugging. Use when the user wants planned iteration changes, bug fixes, or tightly scoped performance edits while preserving the original workflow constraints.
development1
build-plan
Codex wrapper for WF6 architecture planning. Use when the user wants `docs/Implementation_Roadmap.md` and `project_map.json` built from the canonical workflow, templates, and schemas.
development1
baseline-repro
Codex wrapper for WF5 baseline reproduction. Use when the user wants baseline adaptation, reproduction tracking, and `docs/Baseline_Report.md` following the original workflow contract.
development1
final-exp
Codex wrapper for WF9 final experiment planning. Use when the user wants ablations, robustness tests, cross-dataset evaluation, and compute budgeting organized according to the original template.
development1
evaluate
Codex wrapper for experiment analysis and decision-making. Use when the user wants metrics interpreted, a stage or iteration report written, and a NEXT_ROUND, DEBUG, CONTINUE, PIVOT, or ABORT recommendation.
development1
env-setup
Codex wrapper for environment creation and refresh. Use when the user wants the environment detected, created, or synchronized into the legacy `CLAUDE.md` format.
development1
deep-check
Codex wrapper for WF3 deep-check. Use when the user wants a skeptical Go/No-Go review of the technical spec before heavy implementation starts.
development1
orchestrator
Codex wrapper for the canonical WF orchestrator. Use when the user wants project initialization, stage status, gate checks, rollback, or decision logging around `PROJECT_STATE.json`.
development1
auto-iterate-goal
Generate or validate the auto-iterate goal file before launching WF8 auto-iterate
development1
validate-run
Codex wrapper for WF7.5 validation. Use when the user wants the training chain reviewed and smoke-tested before entering WF8.
development1
survey-idea
Codex wrapper for WF1 idea survey and feasibility analysis. Use when the user wants literature-backed validation of a new research idea and a `docs/Feasibility_Report.md` outcome.
development1
code-expert
WF7 Initial Code Generator. Strictly follows project_map.json and Implementation_Roadmap.md to generate all project code in one pass. Used only for initial code generation; subsequent modifications use code-debug.
development1
final-exp
WF9 ablation experiment plan. Design ablation experiments, hyperparameter searches, robustness tests, and cross-dataset evaluations meeting top-venue standards, estimate computation budget, and output Final_Experiment_Matrix.md. Use when main experiments are complete and ablation studies need to be designed.
testing1
env-setup
Environment creation and refresh tool. Create mode sets up a new conda environment; refresh mode detects the current environment and syncs the Environment section of CLAUDE.md. Use when dependencies change or the environment needs initialization.
tools1
.claude/skills/iterate
--- name: iterate description: WF8 structured experiment iteration. Manages the hypothesis→code→run→eval cycle, maintains iteration_log.json, with optional Codex cross-validation. Supported commands: plan (design iteration), code (implement changes), run (execute training + collect metrics), eval (evaluate results), ablate (ablation experiments), status (view progress), log (full history). argument-hint: "[plan|code|run|eval|ablate|status|log] [details]" disable-model-invocation: true allowed-to
development1
validate-run
WF7.5 training pipeline validation. Before entering WF8 iteration, first use Codex to review code for baseline equivalence, then run a 100-step smoke test to verify end-to-end pipeline functionality.
development1
.claude/skills/orchestrator
--- name: orchestrator description: CV research project orchestrator. Coordinates the 10-stage research workflow (WF1-WF10 + WF7.5 gate), tracks progress, and manages PROJECT_STATE.json. Supported commands: init (initialize), status (view state), next (advance stage), rollback (revert), decision (record decisions). Use when the user wants to manage CV research project progress, initialize a project, view status, or switch workflow stages. argument-hint: "[command: init|status|next|rollback|decis
testing1
auto-iterate-goal
Generate or validate the auto-iterate goal file before launching WF8 auto-iterate
development1
baseline-repro
WF5 Baseline Reproduction. Clone comparison method code, adapt to local environment, train and record metrics, output Baseline_Report.md. Used after data preparation and before code planning to provide comparison baselines for the research method.
development1
build-plan
WF6 Code Architecture and Execution Plan. Design the project file structure (separating main research code from reproduced baselines), module pseudocode, configuration schema, and training pipeline. Outputs Implementation_Roadmap.md + project_map.json.
development1
code-debug
Code Fix and Iteration Tool. Handles all code modifications including training error fixes, performance tuning, etc. Can be called by /iterate code or used standalone. After modifying code, creates a semantic commit, then re-trains.
tools1
data-prep
Codex wrapper for WF4 data engineering. Use when the user wants dataset analysis, subset strategy selection, and `docs/Dataset_Stats.md` produced according to the original workflow.
development1
data-prep
WF4 Data engineering and subset generation. Analyzes dataset format and distribution, generates appropriate training subset strategies by project type (NVS/detection/segmentation, etc.), creates data pipeline scripts, and outputs a statistics report.
development1
deep-check
WF3 Second-pass validation and value assessment. Acts as a "devil's advocate" to critically review the technical proposal, search for failure cases, assess risks, and make a Go/No-Go decision. Use after architecture design is complete but before investing in data engineering and coding, to avoid wasting subsequent effort.
development1
evaluate
Result analysis tool (utility). Parses training logs, diagnoses training issues, compares against baseline performance, predicts full-training results, and provides NEXT_ROUND/DEBUG/CONTINUE/PIVOT/ABORT decisions. Can be called by /iterate eval or used standalone.
tools1
init-project
Codex wrapper for staged `CLAUDE.md` generation and updates. Use when the user wants the compact project snapshot initialized or refreshed while preserving the original staged template behavior.
development1
init-project
Project CLAUDE.md phased generator. init mode generates a minimal version (Environment + Workflow), update mode incrementally fills in content (Idea, Structure, etc.) after key stages.
content-media1
iterate
Codex wrapper for WF8 structured iteration. Use when the user wants to run `plan`, `code`, `run`, `eval`, `ablate`, `status`, or `log` while preserving the original iteration schema and workflow logic.
development1
refine-arch
Codex wrapper for WF2 architecture design. Use when the user wants to convert an approved idea into a technical spec, MVP plan, and alternative architecture paths.
development1
refine-arch
WF2 Architecture refinement and MVP design. Reads the feasibility report, analyzes the base codebase architecture, designs plug-and-play new modules, defines the MVP, provides A/B/C alternative plans, and outputs Technical_Spec.md. Use when a research idea needs to be translated into a concrete technical architecture design.
development1
release
WF10 Submission/Release Tool. Multi-scene training, result packaging, filename validation, dry-run submission checks. Used after ablation experiments are complete and before competition submission.
tools1
survey-idea
WF1 Inspiration survey and gap analysis. Takes the user's research idea, performs literature search, gap analysis, competitor analysis, and feasibility scoring, then outputs Feasibility_Report.md. Use when the user has a new CV research idea that needs a feasibility assessment.
business1