EDAgent

Creates and refines academic presentations from research papers. Use this skill for tasks involving the conversion of scientific or technical documents into slide decks for conferences, seminars, lectures, thesis defenses, or academic reviews. Supports beamer LaTeX (.tex), PowerPoint (.pptx), and Markdown output formats.

Execute delay-model validation gates against HPWL baselines. Use when evaluating model consistency, running Gate-0 contract checks, building Gate-1 bucketed scorecards, performing wirelength-aware correlation checks, or deciding readiness for active optimization.

Run a full EDA research chain from knowledge exploration and paper triage to idea debate, hypothesis experiment design, implementation, git versioning, validation, and retrospective. Use when the user requests an end-to-end EDA research workflow rather than a single isolated execution step.

Audit and refine academic manuscripts for structure, terminology consistency, LaTeX hygiene, and high-signal grammar fixes. Use this skill for theses, papers, proposals, and technical reports when the task is manuscript writing rather than experiment execution.

Diagnose and implement backside-routing realization paths, including targeted reroute, local DEF/OpenDB patching, and OpenROAD-backed net-level rerouter bring-up, when theory predicts benefit but the current flow does not realize BM2/BM1 usage.

Audit whether the local BSPDN physical contract is coherent across paper assumptions, GT3 tech collateral, layer/via topology, and current flow policy before promotion or expensive attribution experiments.

Manage and standardize project conda environment usage as an auxiliary utility skill, including env discovery, package availability checks, explicit env selection, and reproducible `conda run -n ...` integration for scripts and workflows.

Explore and structure local EDA knowledge, identify evidence gaps, and prepare targeted literature retrieval tasks for local download and follow-up parsing.

Clean and normalize knowledge-base, tool-registry, and log artifacts by merging duplicates, removing stale items, and correcting inaccurate naming while preserving traceability.

Bridge autoIdea literature recommendation, historical formulation, and idea-generation outputs into the current repo's paper/knowledge/skill workflow.

Retrieve scoped knowledge-base context and tool-registry reuse evidence for an EDA task, then emit compact context artifacts that downstream skills can consume directly.

Extract and maintain reusable experiment knowledge across `log -> result -> conclusion -> experience` layers, then hand that evidence to execution, retrospective, and veto skills without forcing them to re-read raw logs every time.

Maintain and evolve the EDA infrastructure stack (agent policy, knowledge base, tool registry, and skill system) with auditable guardrails and minimal-risk updates.

Enforce experiment hygiene for this EDA repo. Utility skill for knowledge gate, tool reuse checks, and maintenance-log updates during scoped execution.

Capture and maintain reusable script-writing experience across wrappers, helpers, parsers, validators, and runtime shims so future script work can reuse proven patterns and avoid repeated anti-patterns.

Package the agent+skill infrastructure into a standalone, portable repository-ready bundle with skill-local tool mirrors.

Execute one scoped EDA task under governed gates: run bootstrap checks, validate artifacts, and return execution evidence.

Evaluate whether backside benefits come from `CTS-backside-only` or from moving a selected subset of signal nets to backside, under a fair and physically valid comparison contract.

Build and validate theory-grounded optimization models for backside front-vs-back net cost (signal and clock), with strict metric contracts and stability gates against HPWL baselines. Use when users request principled model fitting, crossover reasoning, or promotion decisions from shadow mode to active optimization.

Drive BSPDN objective optimization toward target outcomes (about 8% dynamic-power reduction without area/timing regression, or 5%+ frequency uplift with non-worse power/area) using gated experiments and evidence tracking.

Evaluate whether the current BSPDN PDN contract is strong enough under `BPR reserved for PDN`, separating PDN sufficiency from signal-mixing questions.

Perform post-experiment retrospective for EDA runs, classify failure/success mechanisms, propose high-confidence next actions, and decide whether to recursively trigger a new workflow-scoped-execution iteration. Use after each experiment batch with monitor/summary/manifest artifacts.

Before each experiment, analyze latest route/STA evidence, reflect on method gaps, and output concrete improvement hypotheses and next-step A/B plan. Use when user asks to start new experiments or when model/flow conclusions may be unstable.

Theory-level veto skill for EDA plans. Use before expensive experiment submissions or major flow/model changes to identify logically unsound assumptions, contradiction with known physics/policy, and high-risk invalid comparisons. Produces GO/CONDITIONAL/NO-GO with evidence.

Convert research ideas into falsifiable hypotheses and experiment plans with metrics, controls, pass/fail criteria, and confounder mitigation.

Use git to version-control local research/EDA iterations with reproducible checkpoints, clear commit/tag structure, and explicit per-version differentiation records. Use when users ask to create baselines, compare versions, snapshot experiment states, prepare rollback points, or document each version's characteristics and distinguishing deltas.

Select backside-routing candidate nets after placement/postCTS using power-first criteria with timing-safety vetoes. Use when the task is to build or validate reusable net lists/patterns for targeted GT3 backside reroute, especially when dynamic-power reduction is the primary objective.

Apply and validate GT3 backside routing policy in Innovus. Use when configuring CTS/PDN/signal layer usage, submitting GT3 consistency runs, diagnosing zero-backside-nets, or enforcing BPR-reserved-for-PDN constraints.

Design, adapt, or package synthesizable RTL into meaningful clean testcases or reusable modules. Use when the user wants a new RTL block, wants to trim an existing open-source design into a standalone module, or needs a macro-free testcase with real functionality for synthesis/place/CTS experiments.

Standardize workflow classification and skill selection before execution. Use when a task requires explicit known-vs-unknown workflow judgment, temporary workflow creation, workflow-local skill shortlist construction, or a decision on whether a new skill should be created.

Refines and restructures existing presentations into concise, high-impact versions suitable for short academic conference talks (e.g., 20 minutes). Focuses on logical flow, information density, and visual consistency.

Plan and execute internet-assisted backside net cost modeling for signal and clock nets, then evaluate stability and correlation against HPWL baselines across multiple designs. Use when users ask to improve/validate backside cost models, choose benchmark designs, compare model vs HPWL, or prepare publishable model-evaluation evidence.

Implement EDA research methods from approved hypothesis plans, with integration discipline, measurable contracts, and minimal-risk iteration.

Generate and refine EDA research ideas through structured brainstorming and pro-vs-con debate, then output a testable idea decision memo.

Fetch primary-source paper metadata when evidence is missing or weak for EDA/model/flow claims. Use for requests like building reproducible paper candidate lists, generating user-download queues, and recording citation metadata for local validation.

Summarize local paper PDFs into structured, citation-grounded evidence notes. Use when the user provides local PDF paths and asks for methods/assumptions/results/limitations extraction or wants evidence mapped to current EDA hypotheses.

Create immutable stage checkpoints for Innovus flows. Use when users ask for fixed golden files or restart-from-stage workflows (place/cts/route) with matched DEF+V+SDC and reproducible manifests.