cogni-dao/cogni / .claude/skills/delta-minimizer
delta-minimizer
VerifiedInvestigate ONE specific discrepancy between our wallet's position and a copy-trade target's position on Polymarket. Use whenever the user says 'why are we on the wrong side', 'why is our VWAP off', 'why didn't we mirror that fill', 'delta on market X', 'minimize delta', 'investigate copy-trade gap', '/delta-minimizer', or points at a single market/event on the dashboard and asks what happened. The copy-trade algorithm is known to have fundamental bugs; the skill's job is to find ONE bug, prove it with cited evidence, and stop.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 13.3 KB |
| evals/README.md | markdown | 2.2 KB |
| evals/eval-set.json | config | 852 B |
evals/README.md
View on GitHub# delta-minimizer evals
`eval-set.json` — 9-query set covering the documented trigger phrases and 4 deliberate negatives.
## Bench result (2026-05-07)
Ran via `python3 -m scripts.run_eval` from `.claude/skills/skill-creator/` against:
- `claude-haiku-4-5-20251001` — **4 / 9 passed** (4 negatives correctly suppressed; 5 positives missed).
- `claude-sonnet-4-6` — **4 / 9 passed** (identical pass/fail split as haiku).
Both models correctly suppressed every negative ("promote", "create payments dashboard", "run db migration", "review PR"). Both missed every positive ("minimize our delta against copy targets", "why is our mean delta so high right now", "rank the failure modes…", "loop the delta study every 6 hours", "are we close to swisstony's compounding rate").
## Methodology caveat
`run_eval.py` writes a uniquely-named slash command file (`.claude/commands/delta-minimizer-skill-<hex>.md`) and detects triggering via `tool_use.name == "Skill" || "Read"` containing that unique id. When a skill named `delta-minimizer` already exists at `.claude/skills/delta-minimizer/`, headless `claude -p` resolves the user's query against the _real_ skill (name `delta-minimizer`) rather than the temp command — and the matcher misses, so the eval scores zero triggers regardless of whether the description is correctly leading the model. The 0/5 positive-trigger result therefore reflects an eval-harness collision, not necessarily a description weakness.
## In-session signal
In the active Claude Code session that authored this skill, the harness itself loaded `delta-minimizer` into the available-skills list immediately after `SKILL.md` was written, and trigger phrases like "minimize delta" or "loop the delta study" do route to it. That's the durable signal.
## Vnext for this eval
- Move the real `SKILL.md` to a temp location for the duration of `run_eval`, or
- Patch `run_eval.py` to detect Skill tool calls whose `skill` field equals the _base name_ (not the unique id), or
- Test triggering by reading whether the run produced the documented scorecard markdown shape, not just by tool-call name match.
Either path is in scope of a follow-up `task.NNNN` against `skill-creator`. Not blocking shipping this skill.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |