Overclock is where I collect the skills I build for working with coding agents. They help with specific jobs: remembering a correction, resuming unfinished work, investigating a stubborn bug, handling a review, or writing an explanation someone can follow.
A skill gives the agent a reusable workflow: when to use it, what steps to follow, and what result to produce. Its SKILL.md may be accompanied by scripts, templates, and reference material. A plugin packages one or more related skills for installation.
A skill gives the agent a way to do the work.
Recognize the right task.
Follow a specific workflow.
Produce something you can inspect.
The guides below explain each shipped skill separately. Start with a problem you already have; you can add other packages as they become useful.
Choose your first skill
If you keep repeating corrections, start with Lessons learned. If you lose context between conversations, try Session handoff. For a difficult bug, begin with Debugging discipline.
In Claude Code, add the marketplace once:
/plugin marketplace add luka-zivkovic/overclockEach guide gives the command for its package. If you would like help choosing, install and run the setup advisor:
/plugin install overclock-setup@overclock
/overclock-setup:setupThe advisor checks your current setup and returns recommendations. It does not install the recommended plugins or edit your instruction files.
Skills also ship as directories with their supporting files and Codex metadata. See the repository for the source and current installation details. Keep bundled helpers and references with the skill if you install it separately.
Carry useful knowledge forward
Keep three kinds of knowledge distinct: where the task stands, how the agent should work, and how a project problem was solved.
Different knowledge belongs in different records.
Session handoff
Resume a long task without reconstructing the previous conversation.
session-handoff · session-memory pluginLessons learned
Turn a correction into a reusable lesson, and keep repeated lessons up to date.
lessons-learned · session-memory pluginSolutions
Save a verified fix so the same problem is easier to solve next time.
solutions · session-memory pluginsession-memory contains all three memory skills. learning-loop contains only Lessons learned. They share the same lesson format; install one package to avoid duplicating that skill.
Make decisions with clearer reasoning
Use these skills when the problem, evidence, or language needs clarification before the work can proceed.
Groundwork
Resolve the decisions that matter before planning a substantial change.
groundwork · groundwork pluginCritical thinking
Test a proposal’s assumptions before accepting its conclusion.
critical-thinking · critical-thinking pluginIndependent research
Get a separate, bounded reading of local evidence for a specific question.
independent-research · critical-thinking pluginProject vocabulary
Keep the meaning of your project’s terms consistent as the code and team evolve.
project-vocabulary · project-vocabulary pluginInvestigate and change code
Choose a workflow that matches the engineering problem: a reproducible bug, an unexplained guard, a difficult failure, or a task for another agent.
Test discipline
Demonstrate the bug before changing the code that is supposed to fix it.
test-discipline · discipline-gates pluginGit archaeologist
Find out why a defensive check exists before removing it.
git-archaeologist · discipline-gates pluginDebugging discipline
Build a useful observation loop for a bug that resists straightforward reproduction.
debugging-discipline · debugging-discipline pluginAgent bridge
Ask another installed coding agent for a review or a bounded implementation.
agent-bridge · agent-bridge pluginWork through reviews and writing
Handle feedback and communicate the result. Local review work and publishing a response remain separate steps.
Resolve PR feedback
Turn a GitHub review into justified local fixes and clear draft replies.
resolve-pr-feedback · pr-feedback pluginPublish PR feedback
Publish the exact review replies and thread resolutions you have approved.
publish-pr-feedback · pr-feedback pluginNatural writing
Draft and revise substantial prose in your voice while preserving what it says.
natural-writing · natural-writing pluginChoose a setup and evaluate it
Start with a small useful installation, then add observation and evaluation when you have a question to investigate.
Local eval stack
Connect real agent sessions to a local trace and evaluation workflow.
local-eval-stack · eval-stack pluginSetup advisor
Choose the smallest useful set of Overclock plugins for your work.
setup · overclock-setup pluginThe local evaluation skill connects Ironside, Rubrist, and Casefile. Each product has its own guide and can also be used independently.
What gets captured automatically
After installation, the agent host can select a matching skill from your request or correction. The memory skills can then capture a lesson or a verified fix as part of the work. Because selection depends on the model and context, explicitly ask to save something when you want to be sure it is retained.
The records are plain Markdown under .ai/memory/. Lessons are deduplicated and reinforced, saved fixes can be checked against current code, and handoffs are verified against the live repository on resume.
Overclock doesn’t continuously rewrite your skills. Startup hooks only remind the agent that memory exists. Promoting a recurring lesson into project instructions requires review and approval. Changes to the reusable workflow itself remain changes you can inspect and test.
The repository includes behavioral and routing evaluations that check both useful selection and cases where a skill should stay quiet. Those checks are evidence about particular cases, not a guarantee of the same result in every project. Try a skill on a familiar task, inspect what it changes, and keep the parts that help.