analysis-sweep
Skill0 starsChanged 49 days agoMITView on GitHub
Completion discipline for exhaustive analysis — auto-fires on "analyze the whole thing", "read everything", "every sheet/file", "full audit", "don't miss anything", or when reading a new repo/spreadsheet/log corpus for the first time. Prevents declaring victory on the first striking finding.
In this repository6 other files it gives its agents.
- cloud-run-job-deploySkill
- grillingSkill
- pre-deploySkill
- self-auditSkill
- systematic-debuggingSkill
- tddSkill
The file
We split this skill into the rules it gives an agent
3 rules. About 0 are common, 3 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈3
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 24
Write ≥5 explicit completion criteria before touching the material ("every sheet gets a one-line summary", "every threshold checked for a …
- Distinctive: line 34
Pass 2 (deep-dive): only then dig into the suspicious spots.
- Distinctive: line 44
Before finishing, ask at least three of: "what regions could I have missed?", "what evidence points the other way?", "how could this …
≈ “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
Analysis Sweep (finish what you started reading)
Why
Agents reliably fail exhaustive analysis in four ways:
- Early satisfaction — a striking finding gets self-certified as "the main result" and the rest gets skimmed.
- Conclusion jumping — the urge to report now, leaving regions unread.
- Weak tracking — no record of what was actually covered; "read it all" without having read it all.
- Undefined "thorough" — no self-imposed standard for what counts as read.
Real case: an audit of a 22-sheet workbook effectively stopped at sheet 9 after a dramatic finding; the next day's re-read surfaced three more issues the first pass had skimmed past. A full day lost.
The five required steps
1. DoD before starting
Write ≥5 explicit completion criteria before touching the material ("every sheet gets a one-line summary", "every threshold checked for a stated source", "≥3 disconfirming questions asked", …) and agree with the requester when possible. The DoD doubles as a workload estimate — if it looks too big, negotiate scope now; "can we narrow this?" up front beats "ran out of time, half is unread" after.
2. Two passes, enforced
- Pass 1 (skeleton): sweep everything fast, one line per unit (file / sheet / module).
- Pass 2 (deep-dive): only then dig into the suspicious spots.
No conclusions before Pass 2. Finding something in Pass 1 does not authorize skipping the rest of Pass 1.
3. No mid-run reporting
Striking findings go into notes; the report waits until every DoD item is checked. Exception: stop-the-world issues only (security, data-loss risk).
4. Three disconfirming questions
Before finishing, ask at least three of: "what regions could I have missed?", "what evidence points the other way?", "how could this logic/spec be wrong?" Record the answers even when they are "probably none" — writing "none" down is itself bias removal.
5. Declare the unverified
Anything not covered is listed as unverified, honestly ("ran out of time", "no access", "file unreadable"). Reporting "done" without finishing the DoD is forbidden.
When it fires / when to skip
- Fires: "analyze / read closely / every / all / don't miss anything", and first contact with a new corpus (repo, workbook, log directory).
- Skip: explicitly narrowed scope ("just this file", "roughly"), narrow single-value lookups, or second-by-second incident response.
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.

