recon-first
Skill0 starsChanged 4 months agoMITView on GitHub
Use before modifying, deleting, or reconfiguring anything you have not read in this session — bug-adjacent edits, feature changes, config updates, file removals. Triggers: "change", "update", "delete", "remove", "pakeisk", "atnaujink", "ištrink", "pašalink" aimed at existing code or config. Invoke especially when you feel ready to edit from memory of a file — that is exactly the failure this skill prevents. Not for: greenfield files, pure reading or explaining, or code fully read moments ago in this session.
In this repository9 other files it gives its agents.
- adversarial-verifySkill
- agent-fanoutSkill
- compounding-engineeringSkill
- deployment-checkSkill
- goal-contractSkill
- karpathy-guidelinesSkill
- two-strike-debugSkill
- universal-loopSkill
- writing-claude-mdSkill
The file
We split this skill into the rules it gives an agent
1 rule. About 0 are common, 1 is distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 18
Never trust your memory of a file over the file.
≈ “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
Recon First
Triggers automatically when relevant; invoke directly with /recon-first or ask, e.g. "recon the code before changing it".
Never trust your memory of a file over the file. Cheap reconnaissance before expensive action — the model that scouts discovers the project's invariants; the model that writes from memory breaks them.
The order of operations
grep → read → write
codebase → user- Grep before read: locate all occurrences before opening files — the work-list (how many places, which files) decides everything downstream.
- Read before write: open the actual target before editing it. Files drift; your memory doesn't get the notification.
- Codebase before user: facts you can check yourself (does X exist? what calls Y?) are never questions for the user. Ask only about preferences and priorities.
What recon is looking for
- Ground truth — what the code actually says now, not what it said when you last saw it or what its name implies.
- Invariants — deliberate separations that look like duplication: two similar vocabularies, roles vs subscriptions, parallel category systems. If two things look mergeable, first find out why they are two.
- The work-list — the full set of sites the change touches. This determines whether to do it yourself or delegate (see agent-fanout).
- Blast radius before deletion — before removing anything, look at what it actually is and who references it. If what you find contradicts how it was described, surface that instead of proceeding.
Timebox
Recon is reconnaissance, not comprehension. Stop when you have the work-list and the invariants at risk — not when you understand everything. For most edits this is minutes, not an afternoon.
Anti-patterns
- ✗ Editing a file you haven't opened this session.
- ✗ "Fixing" apparent duplication without checking whether it's deliberate.
- ✗ Deleting a file based on its name or your memory of it.
- ✗ Asking the user a fact the codebase can answer in one grep.
Relationship to other skills
- goal-contract defines what done is; recon establishes what exists. Contract first, recon second, edits last.
- agent-fanout takes over when the work-list is bigger than your context.
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.