# AGILAB

> Reproducible AI/ML workbench for turning notebooks, scripts, and agent runs into executable apps and evidence.

## Agent Skills

AGILAB exposes repo-managed Agent Skills compatible with Codex and Claude Code.
Continue can consume this generated catalog, but AGILAB does not ship a Continue wrapper.
Use `.claude/skills/` as the canonical skill source and `.codex/skills/` as the Codex mirror.
Use `agilab-capabilities.json`, `agilab-capabilities.schema.json`, and `agilab-capability-rules.yml` for the machine-readable inventory of shipped CLI, page, app, package, schema, catalog, and lint-rule surfaces.
Use `python3 tools/agent_context_router.py --files <paths> --prompt "<task>" --json` when an agent needs deterministic AGILAB skill and runbook recommendations before editing.
Use `agenticweb.md` as the compact agentic-web discovery front door generated from the capability manifest.

- advanced-svg-system-design: Design a reusable SVG visual language across multiple related figures. Use when a project needs shared component geometry, connector rules, stable IDs, multi-diagram consistency, and export contracts for web, slides, docs, or reports; use svg-diagrams for a simple new one-off, svg-diagram-tuning for one existing figure repair, and scientific-svg-figures for one publication-grade technical figure. (.claude/skills/advanced-svg-system-design/SKILL.md)
- agilab-code-statistics: Generate fast, reproducible AGILAB code statistics. Use when the user asks for code stats, LOC, file counts, language breakdowns, test/docs/source ratios, churn summaries, or a concise repository size/code footprint report without running builds. (.claude/skills/agilab-code-statistics/SKILL.md)
- agilab-deep-audit: Produce deep AGILAB audit and code-review artifacts with evidence-backed findings, mandatory architecture-foundation readiness, blast-radius tracing, security/test posture, and prioritized recommendations. Use when the user says "review AGILAB", "audit AGILAB", "code review AGILAB", "deep review", "architecture review", "security review", asks for a review document, or asks for comparison-quality critique rather than a quick fix. (.claude/skills/agilab-deep-audit/SKILL.md)
- agilab-docs: Documentation workflow for AGILAB (sources vs generated HTML, public constraints, consistency checks). (.claude/skills/agilab-docs/SKILL.md)
- agilab-evidence-contracts: Maintain AGILAB evidence, proof, replay, and verification contracts. Use when code, docs, tests, or workflows touch run manifests, artifact hashes, first-proof or release-proof evidence, proof capsules, notebook exports, agent-run traces, MLflow handoff, replay commands, or claims about reproducibility and attestation. (.claude/skills/agilab-evidence-contracts/SKILL.md)
- agilab-example-maturity: Improve or review AGILAB packaged examples for external-beta maturity. Use when working on src/agilab/examples, example install/run scripts, example READMEs, app installer example seeding, or tests that enforce example quality, pedagogy, public API usage, deterministic first-run behavior, and newcomer-safe adaptation. (.claude/skills/agilab-example-maturity/SKILL.md)
- agilab-huggingface-spaces: Maintain and deploy the official AGILAB Hugging Face Docker Space using the sibling thales_agilab/huggingface bundle and public agilab checkout. (.claude/skills/agilab-huggingface-spaces/SKILL.md)
- agilab-installer: Guidance for installing AGILAB, installing apps/pages, and debugging install/test failures. (.claude/skills/agilab-installer/SKILL.md)
- agilab-intent-router: Route terse AGILAB operator requests such as "do it", "review AGILAB", "next move", "update repos", "merge it", "check again", "release", and "cluster validation" into the right repo skills, safety mode, validation depth, and output contract using session-derived policy. (.claude/skills/agilab-intent-router/SKILL.md)
- agilab-local-llm: Guidance for using local LLM backends (Ollama/GPT-OSS) inside AGILAB with correctness-first prompts. (.claude/skills/agilab-local-llm/SKILL.md)
- agilab-product-reels: Build and refine short AGILAB product reels and technical demo videos with one-app storytelling, semantic guardrails, frame review, and YouTube packaging. (.claude/skills/agilab-product-reels/SKILL.md)
- agilab-prompt-eval-regression: Design and maintain regression evaluations for AGILAB prompts, local/remote LLM flows, notebook import classification, generated-code routing, agent-skill behavior, and prompt-driven repair or analysis features. Use when a change touches prompt templates, model defaults, local LLM readiness, notebook-to-project import, code generation, or AI-assisted UX. (.claude/skills/agilab-prompt-eval-regression/SKILL.md)
- agilab-pypi-release-maintenance: Guarded AGILAB PyPI release cleanup workflow. Use when an operator needs to inspect, prune, or delete old AGILAB PyPI package releases, especially after a noisy post-release or retention audit item. (.claude/skills/agilab-pypi-release-maintenance/SKILL.md)
- agilab-release-verification: Verify AGILAB release readiness and post-release truth across PyPI, GitHub Releases, release proof, docs, coverage badges, and Hugging Face Space sync. Use when the user asks "ready for release?", "release it", "all good?", "HF aligned?", "why badge failed?", or any release/publication alignment check. (.claude/skills/agilab-release-verification/SKILL.md)
- agilab-runbook: Runbook for working in the AGILab repo (uv, Streamlit, run configs, packaging, troubleshooting). (.claude/skills/agilab-runbook/SKILL.md)
- agilab-security-review-patterns: Review AGILAB changes for security hardening risks. Use when code, docs, or workflows touch installers, Streamlit exposure, cluster/SSH/share behavior, app execution, notebooks, LLM connectors, secrets, PyPI/GitHub/Hugging Face publishing, dependency policy, or external repositories. (.claude/skills/agilab-security-review-patterns/SKILL.md)
- agilab-streamlit-pages: Streamlit page authoring patterns for AGILAB (session_state safety, keys, rerun, UX). (.claude/skills/agilab-streamlit-pages/SKILL.md)
- agilab-testing: Targeted AGILAB validation for unit and app regressions, performance benchmarks and scaling claims, or changes to generated test data and property-test input domains. (.claude/skills/agilab-testing/SKILL.md)
- agilab-ui-robot-validation: Validate AGILAB Streamlit UI changes with the repo's browser and widget robots. Use when touching ABOUT, PROJECT, ORCHESTRATE, ANALYSIS, SETTINGS, sidebar flows, first-proof wizard links, notebook import, screenshots, or public demo UI evidence. (.claude/skills/agilab-ui-robot-validation/SKILL.md)
- chat-export: Export chat transcripts or conversation JSON into clean Markdown, JSON, or plain-text artifacts. Use this skill when a user wants a reusable export of a chat thread, needs a deterministic transcript cleanup, or wants a conversation converted into a file-oriented format for reports, prompts, docs, or downstream tooling. (.claude/skills/chat-export/SKILL.md)
- codex-session-learning: Turn past Codex debugging sessions into reusable bug-fix guidance, prompt templates, and validation rules. Use this skill when the user wants to learn from prior sessions, extract bug/postmortem cases, route future bug logs into stronger prompts, or build an explicit prompt-improvement loop instead of relying on hidden memory. (.claude/skills/codex-session-learning/SKILL.md)
- docs-publish-github-pages: Review, fix, and validate docs publishing workflows for GitHub Pages. Use this skill when a user needs a docs deploy workflow corrected, wants versioning behavior explained, needs publish triggers adjusted, or wants a Pages deployment flow verified end to end. (.claude/skills/docs-publish-github-pages/SKILL.md)
- docx-figure-sync: Replace or update embedded figures in DOCX reports while preserving placement, sizing, and nearby captions. Use this skill when a user wants a figure updated inside a Word document, wants SVG/PNG media replaced without rebuilding the whole document manually, or needs related PDF regeneration only when explicitly requested. (.claude/skills/docx-figure-sync/SKILL.md)
- notebook-to-agilab-project: Migrate or maintain a small local notebook workflow inside an AGILAB project. Use this skill when a user wants notebooks turned into a reproducible AGILAB project, project-owned notebooks exposed under ANALYSIS, WORKFLOW notebook import, lab_stages.toml, artifact contracts, and a conceptual workflow view. (.claude/skills/notebook-to-agilab-project/SKILL.md)
- pipeline-concept-view: Add or refine a conceptual workflow view alongside a generated execution view without hard-coding app semantics into the generic UI. Use this skill when a user wants a pipeline_view.dot/json file, a conceptual architecture diagram, or a lab_stages.toml reviewed for clearer naming, IO flow, and semantic alignment with the WORKFLOW page. (.claude/skills/pipeline-concept-view/SKILL.md)
- plan-before-code: Plan before editing code. Use this skill for multi-step implementation, debugging, refactors, release work, or any coding task where sequencing, assumptions, or validation matter. It enforces a short plan first, validates assumptions before edits, and verifies results before close-out. (.claude/skills/plan-before-code/SKILL.md)
- repo-skill-maintenance: Maintain repo-managed agent skills across `.claude/skills` and `.codex/skills`, including targeted sync, validation, index regeneration, drift checks, and Tokki skill visibility. Use when adding or updating a shared skill, migrating a user-managed skill into the repo, or reconciling agent skill copies without overwriting unrelated skills. (.claude/skills/repo-skill-maintenance/SKILL.md)
- report-qa-docx: Review DOCX reports for missing figures, stale wording, duplicate sections, caption drift, and local consistency issues. Use this skill when a user asks for a quality pass on a Word report, wants to compare exported versus source versions, or needs a review-driven cleanup checklist before delivery. (.claude/skills/report-qa-docx/SKILL.md)
- scientific-svg-figures: Create or substantially redesign one publication-grade scientific or technical SVG for reports, slides, README/docs, or DOCX/PDF workflows. Use for architecture, methodology or training-loop, pipeline, comparison, result-summary, timeline, and annotated-system figures whose deterministic cross-medium layout is part of the deliverable; use svg-diagrams for a simple one-off, svg-diagram-tuning for a small repair, and advanced-svg-system-design for a reusable figure family. (.claude/skills/scientific-svg-figures/SKILL.md)
- slides: Create, edit, inspect, or explain presentation slide decks (`.pptx`) with PptxGenJS, bundled layout helpers, and render/validation utilities. Use when tasks involve building a new PowerPoint deck, recreating slides from screenshots/PDFs/reference decks, modifying content while preserving editable output, extracting or explaining a specific slide, adding charts/diagrams/visuals, or diagnosing layout issues such as overflow, overlaps, and font substitution. (.claude/skills/slides/SKILL.md)
- slides-docx-align: Align a slide deck with a DOCX report while preserving each artifact’s role. Use this skill when a user wants missing report figures inserted into slides, terminology synchronized across report and deck, or extra slides added without overwriting the current summary version. (.claude/skills/slides-docx-align/SKILL.md)
- svg-diagram-tuning: Repair one existing SVG without redesigning its meaning. Use for targeted readability defects such as cropped text, crowded blocks, stale connector anchors, unclear arrows, or overlapping labels; use svg-diagrams to create a simple new figure, scientific-svg-figures for substantial publication-grade figure work, and advanced-svg-system-design for a reusable figure family. (.claude/skills/svg-diagram-tuning/SKILL.md)
- svg-diagrams: Create a simple standalone repo-native SVG or directly edit its boxes, text, and connectors. Use for one-off architecture sketches and explanatory visuals; use svg-diagram-tuning for repair of an existing crowded figure, scientific-svg-figures for publication-grade technical figures, and advanced-svg-system-design for a reusable multi-figure visual system. (.claude/skills/svg-diagrams/SKILL.md)
