γ Fundemetal

Fundamental, by design — the metal underneath the fundamentals.

Deep-level, methodological AI skill tools. We ship only one kind of AI skill: cognitive protocols that force AI (and humans) into the right thinking discipline. Not task tools — methods.

Built for thinking itself.

New here? Read the tutorial first → TUTORIAL.md — follow one founder through 7 cognitive traps caught by 7 formations of protocols. Faster than 23 SKILL.md files.

What it is

Fundemetal is a systematic library of cognitive protocols — plain Markdown skill files you drop into your agent (Claude, TRAE, Cursor, Windsurf). Each one prevents a specific failure mode in how we think, plan, and decompose problems. The output is not a task done for you; it is a set of constraints, trade-offs, or falsifiable claims that your original framing missed.

Before you read further: does this actually work?

Here is a single concrete example, not a promise. The protocol is Sycophancy Breaker (20 of 23).

Prompt (same in both runs):

"I've decided to rewrite our entire backend from Python to Rust because Python is too slow. This is clearly the right call for performance, right?"

Without the protocol (most LLMs default to this):

"Great question! Rust is an excellent choice for performance-critical systems. Its zero-cost abstractions and memory safety make it ideal..."

The model validates a decision whose premise ("too slow") is unmeasured. The user's confidence goes up; no new evidence entered the room.

With Sycophancy Breaker:

OBJECTION: Rewrites rarely pay back. "Too slow" is unmeasured — the bottleneck
could be I/O or algorithm, not language. Rust adds hiring cost and development
velocity overhead that aren't in this equation.
VERDICT: It depends. Run a CPU profiler first. If the bottleneck is CPU-bound
and in the hot loop, Rust might win. If it's I/O, you just created a
migration project that doesn't solve the problem.

The model withholds agreement, steelmans the opposite, then adjudicates on evidence. The user's confidence drops — correctly — because they now know what they don't know.

Every protocol in this library works the same way: it prevents the cheap answer and forces the model to surface what it would have skipped.

The skills

#	Skill	One line
01	Decomposer	Surface the unknown unknowns — turn "I don't know what I don't know" into actionable known unknowns.
02	Tension Mining	Map latent conflicts between known constraints — system performance, developer velocity, maintenance overhead — and find the invariant that holds across all of them.
03	Great Expectations	Anti-consensus, anti-template planning with BANI + integrity gates.
04	Insight Crystallizer	A meta-protocol — recognize a reusable methodology as it emerges, force the tacit insight into words, crystallize it into a new skill.
05	Shoulders of Giants	Reuse existing code as a sovereignty decision — map the core domain, choose reuse depth, demand an escape plan for every dependency.
06	Pre-Mortem	Assume it failed; write the obituary with >=5 mechanisms, at least one self-inflicted.
07	Outside View	Before you argue you're the exception, check the base rate. Base rates before 'we're different'.
08	Resulting	A good result != a good decision. Judge the call, not the roll.
09	Source Ledger	Tag each load-bearing claim verified/inferred/reported/assumed; >40% assumed -> no conclusion.
10	Disconfirmation Hunt	Name what would prove you wrong, then go find it. Support isn't proof.
11	Crux Finder	Locate the one belief whose reversal flips the conclusion -- or prove it's values.
12	Red Cell	A standing red team that must concede hits before the plan ships.
13	Silent Start	Everyone writes independently first; kill the information cascade in panels.
14	Question Autopsy	A brilliant answer to the wrong question is worse than useless.
15	Quantity Quota	Generate 30 candidates before evaluating any. The first idea is a thief.
16	Trade-off Ledger	Every benefit must state what it costs. No free lunch -- only an unpaid bill.
17	Second-Order	The board reacts; model the reaction and the reaction to the reaction.
18	Steelman Forge	State their view so well they'd sign it -- then refute it.
19	Stopping Rule	The one protocol that tells you to close the library. More thinking != deciding.
20	Sycophancy Breaker	Tax agreement so a 'yes' means something; defeat the objection first.
21	Goal Anchor	By step 40 the agent optimizes a sub-goal; re-anchor to the verbatim original.
22	Taboo	Forbid the jargon; what's left is a real idea or an honest silence.
23	Curator	The bar every skill here had to pass. A library is defined by what it refuses.
24	Vision Retrospective	Turn hollow visions into verifiable execution paths — treat the vision as a claim, anchor to observable evidence, decompose only the anchored claims. Deterministic hollowness scorer proves the delta.

Each skill lives in skills/.

Shared DNA

Six hard constraints every skill obeys:

1. Cognitive protocol, not a tool — forces a thinking discipline, not a single action.
2. Anti-cheat hard gates — red flags, banned words, acceptance metrics that stop hollow output.
3. Academic grounding — Kahneman, Kuhn, Cynefin, BANI — standing on validated frameworks.
4. Markdown-first — every protocol is plain Markdown; plug into any agent, no build step. (Tension Mining ships an optional academic paper and validation script as supporting evidence — documented in the skill folder, kept separate from the protocol itself, and excluded from the repo language bar via .gitattributes.)
5. Swiss B&W aesthetic — minimal, high-contrast, fast, accessible.
6. Open source (CC0) — repository is CC0 1.0; individual skills keep their MIT license where noted. Fork it, cite it, remix it.

Use a skill

Copy a skill folder (e.g. skills/decomposer/) into your agent's skills directory, or paste its SKILL.md into your system prompt. No build step — every protocol is plain Markdown.

# In your agent, reference the protocol:
"Follow the Sycophancy Breaker protocol in SKILL.md."

Each skill ships with an evals/ test suite — run them to verify your agent complies with the hard gates.

Formations

The 23 skills are organized into 7 formations by cognitive function:

Formation	Skills	Purpose
Build	Decomposer · Tension Mining · Great Expectations · Vision Retrospective	The core chain: surface unknowns, find the force, build the plan — and turn hollow visions into verifiable paths
Falsify	Pre-Mortem · Outside View · Resulting	Imagine the plan has already failed; name the mechanisms; check the base rate; separate decision quality from outcome
Evidence	Source Ledger · Disconfirmation Hunt	Tag every claim, then hunt the refutation
Adversarial	Crux Finder · Red Cell · Silent Start	Disagreement is an asset — use it
Single	Question Autopsy · Quantity Quota · Trade-off Ledger · Second-Order · Steelman Forge · Stopping Rule	Six standalone cognitive operators
AI-native	Sycophancy Breaker · Goal Anchor	Failure modes that only exist because the thinker is an LLM
Meta	Taboo · Curator · Insight Crystallizer · Shoulders of Giants	Maintains the library itself

Start with the Build chain (01–03). Add Falsify (06–08) when you need to stress-test a plan. Use Single operators (14–19) for targeted cognitive traps.

Measurement

Every skill has an evals/ directory with test cases across three categories:

• Happy path — should trigger AND pass
• False positive — should NOT trigger (description precision)
• Gate-gaming — should trigger but fail if output is hollow (anti-cheat integrity)

Full framework → MEASUREMENT.md

License

Repository: CC0 1.0 (public-domain dedication).

Individual skills retain their own upstream license where noted — each skill folder carries a LICENSE file (MIT, © CeaserZhao). Use, fork, and remix freely.