odb-debug

Ask a placed or routed design questions through a persistent OpenROAD session on a flow stage's ODB (the odb-debug daemon and its MCP tools): where the macros are, which cells sit inside a macro footprint or in a sliver between macros, what check_placement and the worst paths say. Use when a legaliser fails or stalls, when a floorplan looks wrong, or whenever the answer is in the database rather than the log.

Jump to install

Repository: The-OpenROAD-Project/bazel-orfs
Last source activity: September 18, 2026 at 06:45
Detected SKILL.md language: English
Stars: 27
Forks: 24

Install options

The review-first prompt is selected by default. You can switch to a direct command or download a local copy.

Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.

A direct command skips the review prompt. Inspect the source before running it.

npx skills add https://github.com/The-OpenROAD-Project/bazel-orfs --skill odb-debug

The command stays on one line. Scroll horizontally to inspect it before copying.

Prefer a local copy? Download the files currently available to SkillsMP.

Review the source files

Read SKILL.md and any companion files shown by SkillsMP before deciding whether to install.

SKILL.md

Source instructions · Read-only preview

name: odb-debug
description: Ask a placed or routed design questions through a persistent OpenROAD session on a flow stage's ODB (the odb-debug daemon and its MCP tools): where the macros are, which cells sit inside a macro footprint or in a sliver between macros, what check_placement and the worst paths say. Use when a legaliser fails or stalls, when a floorplan looks wrong, or whenever the answer is in the database rather than the log.

# odb-debug The instructions for this skill are maintained in the shared Claude slash command. Please read and follow the single source of truth: [../../../.claude/commands/odb-debug.md](../../../.claude/commands/odb-debug.md)

View on GitHub

Back to search Back to installation

repair-timing-grt

Investigate OpenROAD global-route repair_timing correctly — the grt "grind"/timeout, hold- and setup-buffer insertion. Covers why a skip-repair (fast) grt measures nothing about repair, why repair_timing must run inside global_route.tcl on the pre-GR ODB, how to split the grind into setup vs hold before blaming either, why the win is measured as clk_period-WNS in picoseconds rather than as a percentage of WNS, and why a design that closes measures nothing about a repair knob. Use when a grt stage times out or runs for hours, when a change is meant to affect repair_timing, or when reasoning about hold vs setup buffer insertion at global route.

2026-09-27

cache-miss

Ask whether a target is a remote cache hit on this machine without building it (a capture kills every miss as it appears and lists the actions behind it as not reached, minutes on a laptop), and find out why it is not when another machine built it: a few kilobytes of key evidence per machine (Bazel version, commit and tree, redacted options, tool digests, the source files read, one line per action with its cache key and which of arguments, environment, tools, sources or design inputs moved), committed next to the design and diffed, naming the first action that differs and the source file behind it. Says what committed evidence promises and when to recapture. Use before building any long flow target, when XSTile_grt or any stage rebuilds unexpectedly, or before blaming Scala, a toolchain or the cache.

2026-09-26

dogfooding

The standing rules learned building XiangShan with bazel-orfs, for any large design: fix the tools and never the design (no multicycle or false-path exceptions, no added RTL registers); hard stops where a problem is made; finish the macros before the parent; a design README says what is; public PR tests check intended behaviour, not golden .ok output; a PR on an unmerged upstream PR is a draft; choices in the flow script, facts in the .odb; relative paths; unattended means continue; Chesterton's fence, check what our flow skipped before blaming a tool; watch a run that grinds past its expected time. Use before proposing a fix for a timing or flow problem found on a big design, before writing a design README or status, before opening a PR that depends on another, and before writing a test for a public PR.

2026-09-25

View all from this repository

odb-debug

Source facts

Install options

Review the source files

More from this repository

More from this repository