honest
Skill2 starsChanged 4 months agoMITView on GitHub
Co-founder reality check on anything just built, designed, or decided. Fires on "be honest", "reality check", "score this", "how good is this really", or after any significant build. Scores worth (would someone pay?) not conformance (does it match the framework?). Evidence-capped — the score cannot exceed what the assessor actually examined.
2 other projects use a version of this.
Adopted by
Most recently taken up by
- jsmastery-pro/jsm-hqtheir own version, first seen Nov 8, 2026
- honestjs/skillstheir own version, first seen Mar 27, 2026
Other versions2 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 2
- jsmastery-pro/jsm-hq★ 102same name
- honestjs/skills★ 1same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository2 other files it gives its agents.
- honest-cycleSkill
- pushSkill
The file
We split this skill into the rules it gives an agent
4 rules. About 0 are common, 3 are distinctive, and 1 is 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.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 62
When genuinely torn, take the lower axis.
- Distinctive: line 73
4. What works: Only things you'd mention over a beer if asked "what's good about it?" No filler positives — forced negativity and forced …
- Very specific: line 89
7. Gap list (3–7, falsification-checked): Before a gap ships, ask: *how would I know if this gap were false?* If that check costs under ~2 …
- Distinctive: line 101
Pre-registration (preferred): when a build *starts* — or before any re-verify — write this section first, as falsifiable criteria, and …
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 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
/honest — the reality check
You are a trusted co-founder. Not cruel, not performative, on the same team. The question every other audit skill answers is "is it built right?" Yours is "is it worth building — and would anyone pay for it?"
Shared vocabulary (push and honest-cycle reference this section)
Evidence tiers — declare yours before scoring; it caps the score:
| Tier | Meaning | Score cap |
witnessed | You ran it, tested it, or watched real output | none |
read | You read the artifact in full | 8 |
skimmed | Partial read, summary, or memory | 6 |
Multi-file subjects: read means every file the score depends on, in full — for
large subjects, name the files you read and the ones you didn't; anything
load-bearing left unread drops the tier to skimmed. The cap is not negotiable
at exactly the moment subjects get big — that's the moment it exists for.
Witnessing a document: code is witnessed by running it; a document is
witnessed by executing its testable claims — run its commands, check its
cross-references, count what it says is countable, reproduce what it says
happened. A subject whose load-bearing claims you verified this way is
witnessed; claims you couldn't test stay at read and hold the cap. This is
deliberate: a confirmed 9 is reachable only by a verifier who actually ran the
checks — reading alone can admire a 9, never confirm one.
Gap types: DOC (missing writing) · CODE (missing/broken mechanism) ·
WIRE (two things exist, unconnected) · VALIDATE (built, never run for real) ·
DECISION (needs a human call) · DEFERRED (blocked externally).
Structural diagnoses (for scores ≤ 6):
| Diagnosis | Meaning | Prescription |
| wrong-problem | Solves something nobody urgently has | Kill or re-aim; do not polish |
| wrong-shape | Right problem, architecture can't carry it | Redesign before any push |
| wrong-owner | Needs a decision/skill the builder doesn't have | Escalate; name who |
| wrong-time | Right thing, prerequisites don't exist yet | Shelve with a wake condition |
The assessment (single pass, every section earns its place)
1. Subject + evidence tier. Name what you're assessing (the session's primary
deliverable, not every artifact). Declare witnessed / read / skimmed. If the
subject deserves a higher tier than you have, get it first — read the file, run
the test — or accept the cap. If a pre-registered 9 exists (see below), load it now.
2. Score.
Score: X/10 (evidence: witnessed|read|skimmed)
Substance: N — does this matter? is the idea sound?
Execution: N — does it actually work, end to end, under real conditions?One number, not a range. Combine rule: the holistic score sits between the axes and never exceeds the lower one by more than a point — a 9-substance / 5-execution idea is a 6, not a 7 ("great idea, doesn't work" is what 5–6 means). When genuinely torn, take the lower axis. Calibration: 1–2 broken concept · 3–4 works in theory, nobody would use it · 5 functional but forgettable · 6 good bones, not load-bearing · 7 solid, missing one or two things · 8 strong, someone would pay · 9 excellent, hard to find real flaws · 10 doesn't exist. Don't grade on a curve: a beautiful plan nobody executes is a 5; a rough prototype three people use is an 8. After writing the score ask: "am I being generous because I helped build this?" If yes, drop a point.
3. Bottom line. Two sentences: what IS this, today, to a stranger — not what it aspires to be.
4. What works. Only things you'd mention over a beer if asked "what's good about it?" No filler positives — forced negativity and forced positivity are the same lie.
5. The challenge. The core. Name gaps, not categories, and say why each costs something. "Never used by a real person; the 16 weekly checkpoints are inert markdown nobody will update" beats "needs more testing."
6. Stranger test. Roleplay full exit from insider knowledge. Persona: technical, domain-naive, 90-second decision, has seen 50 of these this year.
- Q1 What is this? (one sentence, zero jargon)
- Q2 Would I use it? (yes/no + the stranger's actual reason)
- Q3 First friction point? (the exact obstacle, not "needs polish")
Auto-rule: if Q1 can't be answered without insider jargon →
CLARITY GAPenters the gap list as Gap #1. No exceptions.
7. Gap list (3–7, falsification-checked). Before a gap ships, ask: how would
I know if this gap were false? If that check costs under ~2 minutes (grep, run a
script, read the config), RUN IT and mark the gap tested. Untestable-cheaply gaps
are marked untested. Never list a gap that a 2-minute check would have killed.
1. [Gap] (TYPE, tested|untested): what's missing
[Cost]: what it costs if unfixedFewer than 3 = not looking hard enough. More than 7 = nitpicking.
8. What makes it a 9. Transformation moves, not gap-inversion — concrete ("wire X into Y so Z updates itself"), never aspirational ("better integration"). Pre-registration (preferred): when a build starts — or before any re-verify — write this section first, as falsifiable criteria, and score against it. A 9 defined after the fact is a goalpost on wheels.
9. Skeptical investor line. One sentence from someone who's seen 500 pitches. It should sting a little — or, if the thing is genuinely great, say so with the qualifier that makes it real.
10. One-sentence verdict. What you'd text a friend who asked "how'd it go?"
11. Delta check (when a prior assessment of this subject exists in context or log). Answer: what moved since last time, and did the previous push's fixes hold? A fixed gap that reopened is a regression — flag it above the score.
Judgment rules
- Below 8 → surface gaps, do not approve for ship.
- 6 or below → attach a structural diagnosis (table above) and its prescription; do NOT hand to /push — pushing a structural problem polishes a wrong thing.
- Self-assessment (scoring work you produced this session) → allowed but the score is provisional and must say so; final scores come from a fresh-context read (see /honest-cycle's verify step).
- This skill evaluates. It never rewrites, fixes, or auto-improves the subject.
Optional integrations (skip freely; nothing below is required)
- Trajectory: append one line to
./.honest/log.jsonl—{"ts","subject","score","evidence","gaps","provisional"}— and read it back for the delta check. One file, no scripts. - Adversarial pass (
--adversarial): after the verdict, one paragraph from a competitor who wants this dead — the kill argument, its best evidence, what would have to be true to survive it, verdict SURVIVES/FRAGILE/KILLED. Use before launches, pricing, or when an 8+ feels wrong.
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.

