Skip to content
View decentralizeallengineering's full-sized avatar
  • Joined Oct 27, 2026

Block or report decentralizeallengineering

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse

decentralizeallengineering

Small, tested tools for working with coding agents (Claude Code mostly), office files and a homelab LLM rig. Built while developing, used every day. Most are one standard-library Python package; all are Apache-2.0, have CI and a tagged release.

Start here

repo what it does try it
codeindex Local code index for agents: where a symbol is defined, who calls it, which tests cover it. Skill plus MCP server. claude plugin install codeindex@decentralizeallengineering
zoomout When an agent is stuck, restate the goal and choose: continue, pivot or simplify. claude plugin install zoomout@decentralizeallengineering
pacekeeper Claude Code usage pacing from the real 5h and 7d meters: targets, spend by project, a wind-down hook. pip install git+https://github.com/decentralizeallengineering/pacekeeper

The two plugin lines need this marketplace once:

claude plugin marketplace add decentralizeallengineering/decentralizeallengineering

Python tools install with pip install git+https://github.com/decentralizeallengineering/<repo>; single-file tools say so in their README.

Agent quality and evals

repo what it does
donegate Stop hook: no "done" unless a test ran after the last edit.
taskmine Mine benchmark tasks from git history: tests red at the parent, green at the fix.
toolcensus Count an agent's real tool calls; check them against another tool set (MCP, OpenAI, Anthropic).
freshlook Independent audit of a repo by a clean agent that sees only code and intent.
anti-lobotomy Audit a Claude Code setup for silent context loss.
intentpin Pin your own words into every session start and compaction.
breakthroughs Several cheap agents brainstorm; one stronger agent scores the ideas.
oldest-asks Find what you asked your agents for that was never delivered, oldest first.

Agent cost

repo what it does
turnthrift Fewer turns: run many reads in one call; hooks nudge batching and short replies.
leanagents Route subagents to small-prompt types and cheaper models.
autocontext Pick context window and autocompact threshold from measured cost.
hushwire Cut cross-session agent messages: status-only gate, digests, stats.
escalator Cheap-first patches: a local model drafts, a stronger model fixes on red.
qsum Extractive digests of long files through a local OpenAI-compatible endpoint.

Many sessions at once

repo what it does
fleetwarden Guardrail hooks and a launcher: protected dirs, rm and wait guards, agent routing.
supra-orchestrator One orchestrator for many sessions: order ledger, cadence, escalation.
landbot Local merge queue for agent branches: batch, gate, bisect, land.
billet Hand a long session over to a fresh successor with a checked brief.
tickwise Adaptive tick intervals: slow when nothing changes, fast on change.
diplomat One idle watcher session per machine: alarms, load, tmux keepalive.
opsledger Split a big order into deduped asks with evidence-backed status.

Office files

repo what it does
xlsxops Excel operations for LLM tool calls, with JSON schemas and safe saves.
docxops Word operations for LLM tool calls, including a real table of contents.
pdfops PDF text, tables, merge, split and form fill for LLM tool calls.
officecheck Grade .docx, .xlsx, .pptx and .pdf outputs against a JSON spec, no model judge.
fakepii-docs Fake-PII office documents with ground truth, to test maskers.

Homelab LLM rig

repo what it does
llm-rig Local LLM servers in libvirt VMs with GPU passthrough.
gpu-handoff Move a GPU between a VM and the host without a reboot.
vmlease Lease-based VM orchestration; a reaper reclaims.
modelstage Staged GGUF model rollout with health gate and rollback.
gsched GPU airtime scheduler with leases.
jobq Heavy jobs in cgroup slices, one queue across machines.
rigwatch GPU and VM safety: temperature, RAM and job guards.
nascare NAS care: heat, spin-down, SMART, ZFS error growth.
corsair-fans Temperature-driven fans for a Corsair Commander Core on Linux.

Small utilities

repo what it does
trash30 A trash folder that empties itself after 30 days.
addonly Put a file into a folder you only ever add to; never overwrites.
uistrings Pull UI strings from HTML and JS; flag filler, emoji, long labels.

Issues and pull requests are welcome on each repo. License: Apache-2.0.

Popular repositories Loading

  1. landbot landbot Public

    Local merge queue for AI-agent branches: batch, gate, bisect, land. No CI service needed.

    Python

  2. fleetwarden fleetwarden Public

    Guardrail hooks and a session launcher for running many Claude Code sessions: cadence, agent routing, protected dirs, rm and wait guards, intent injection.

    Python

  3. codeindex codeindex Public

    Local code index for coding agents: where is it defined, who calls it, which tests cover it. No LLM, no network, MCP server included.

    Python

  4. qsum qsum Public

    Extractive digests of long files through a local OpenAI-compatible endpoint.

    Python

  5. opsledger opsledger Public

    Turn a big multi-part order into deduped, deconflicted asks with evidence-backed status and a percent-complete ledger.

    Python

  6. anti-lobotomy anti-lobotomy Public

    Audit a Claude Code setup for silent context loss: compactions, truncated reads, silent model rewrites, rule bloat and alarm noise.

    Python