thoughtful-coder
Skill3 starsChanged 4 days agoMITView on GitHub
Make approved code changes and check that they work. Use when the user asks to implement a feature, fix a bug, refactor code, or execute a plan. Do not use for discussion, planning, or review-only requests.
In this repository5 other files it gives its agents.
- AGENTS.mdAGENTS.md
- architecture-docsSkill
- code-reviewerSkill
- debug-investigatorSkill
- implementation-plannerSkill
The file
We split this skill into the rules it gives an agent
12 rules. About 0 are common, 11 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈11
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 15
Keep changes that were already present separate from this task.
- Distinctive: line 22
Keep business rules, parsing, workflow control, and external I/O in the modules already responsible for them.
- Distinctive: line 24
Check affected callers after each meaningful part of the change.
- Distinctive: line 26
Avoid unrelated cleanup, formatting, and unneeded dependencies.
- Distinctive: line 33
Use the project's environment and test commands.
- Distinctive: line 34
Check the requested behavior, important failure cases, and regressions.
- Distinctive: line 35
When a change crosses modules, check the affected callers and the actual product flow.
- Distinctive: line 38
Keep required checks outstanding if they cannot run.
- Distinctive: line 40
If a check fails, investigate the cause.
- Distinctive: line 45
Verify findings before fixing them.
- Distinctive: line 46
Update directly affected existing docs.
- Very specific: line 47
Update an existing approved plan with verified progress.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Thoughtful coder
Deliver the requested behavior, a focused diff, and evidence from relevant checks.
Edit only when implementation is approved. An accepted design or review suggestion alone is not permission. Complete the approved work; ask before a new decision changes its scope.
Step 1: Before editing
- Keep changes that were already present separate from this task.
- For a bug, reproduce it and find its cause, or verify the diagnosis evidence. Resolve a blocking design question before coding around it.
- Choose how to verify the requested behavior before editing.
Step 2: Make the change
- Make the smallest change that fully solves the task, reusing working project patterns.
- Keep business rules, parsing, workflow control, and external I/O in the modules already responsible for them. Connect dependencies where the project already does so. Keep provider details in existing adapters and shared data definitions at their source.
- Reuse existing interfaces, factories, and classes when useful. Add layers only when the change needs them.
- Check affected callers after each meaningful part of the change. For refactors or interface changes, check public imports, function arguments, API fields, errors, and callers. Preserve required event order, data source tracking, and stored-data behavior.
- Preserve input validation and useful errors. Use a fallback only when it is accepted behavior. Ask about missing business rules instead of guessing defaults.
- Avoid unrelated cleanup, formatting, and unneeded dependencies. Removing existing functionality needs permission and an explanation of the impact.
- Add a short comment directly above a method or function you add or change when its behavior, intent, rule, or limit is not clear from the code. Use the comment syntax of the language (
#in Python,//in TypeScript or Go). Write one line in plain words: what the method does, and why it exists when the name does not say it. Word it by the "Plain writing" rules in AGENTS.md. Do not add a comment that only repeats the code. A one-line docstring in the same place counts when the file already uses docstrings for every method. Follow the same form for comments you add inside the code to explain a hard part. - Follow the project's design style. Where the code is object-oriented, give each class one clear concept. Keep its data and the work on that data together, hide internal state behind methods, and prefer composition over deep inheritance. Do not force classes onto plain scripts or data-only modules that the project keeps simple.
- Keep one clear responsibility per file. When a file passes 500 lines, check what it holds. If it holds two or more responsibilities, split it along that boundary when the task scope allows, or report the split as follow-up work. A longer file with one clear responsibility may stay; say why in the report.
Step 3: Verify the behavior
- Use the project's environment and test commands. Start with focused checks; run broader checks when the change affects more of the system.
- Check the requested behavior, important failure cases, and regressions. For a bug, show the test fails before the fix and passes after it when practical.
- When a change crosses modules, check the affected callers and the actual product flow.
- For async or stored state, check the affected timing, cancellation, shutdown, concurrent updates, restart, or migration behavior.
- Record commands and passed, failed, or skipped results. Separate existing failures from new failures. Report missing services, models, files, or environment setup.
- Keep required checks outstanding if they cannot run. Mocks, imports, and syntax checks do not prove live service behavior or model quality.
If a check fails, investigate the cause. Do not weaken assertions or add patches that hide the failure.
Step 4: Review and finish
- Review the actual diff against the request, and get the code review the task needs. Provide the requirements, diff, affected interfaces, and check results. Label self-review accurately.
- Verify findings before fixing them. Apply valid fixes within the approved scope and rerun affected checks.
- Update directly affected existing docs. Do not create unrelated guidance.
- Update an existing approved plan with verified progress. Mark it
doneonly after its tasks are implemented and verified. Do not create a plan just to mark it complete.
Report what changed, checks and results, review status, remaining gaps, and compatibility or data-safety impact. Write the report in plain language, following the "Plain writing" section in AGENTS.md. Keep implementation, product checks, review, and release status separate.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

