Agent skill

Pick Invariant

by cyfung1031 in cyfung1031/userscript-supports

Default first-line, domain-agnostic meta-reasoning control for choosing the right path.

MITAuto-check passed

Install Pick Invariant

skills CLI
$ npx skills add cyfung1031/userscript-supports --skill pick-invariant -a claude-code

Project install by default; add -g for ~/.claude/skills/.

GitHub CLI
$ gh skill install cyfung1031/userscript-supports pick-invariant --agent claude-code

Project scope by default; add --scope user for a personal install. Needs GitHub CLI 2.90.0 or later (public preview).

Manual copy
$ git clone --depth 1 https://github.com/cyfung1031/userscript-supports.git skills-src && mkdir -p .claude/skills && cp -r skills-src/agent-skills/pick-invariant .claude/skills/pick-invariant && rm -rf skills-src

Use ~/.claude/skills/ instead of .claude/skills for a personal install. The folder must contain SKILL.md.

Claude Code skills documentation · loads skills from .claude/skills/

Facts

Skill name
pick-invariant
GitHub stars
121
Token cost
~2.2k tokens
SKILL.md length
769 words
Files
56 (incl. references)
Skills in repo
6
Repo updated
First seen
Licence
MIT

At a glance

Default first-line, domain-agnostic meta-reasoning control for choosing the right path.

  • Works in 6 steps: Authority and mode → Discovery, depth, and presentation → Audit and structural commitment → …
  • Resolve an undecided choice
  • SKILL.md covers 1. Authority and mode, 2. Discovery, depth, and…, 3. Audit and structural… and 4. Hard specialist firewalls, plus 2 more sections
  • Instructions only: no scripts, shell commands, URLs or credentials in SKILL.md

What it does

Pick Invariant is an agent skill from cyfung1031/userscript-supports. Default first-line, domain-agnostic meta-reasoning control for choosing the right path. Use it to resolve an undecided choice, narrow an exploration, audit an invariant, investigate an owner-identified gap, or determine whether a validated owner can be used directly when no Pick exploration or audit is requested. Preserve target, authority, and semantics; retain only observable distinctions that can change the target decision; prefer the cheapest sufficient observation or representation; refine only on an…

Its SKILL.md is about 2.2k tokens, which your agent loads only when the skill is triggered. The skill folder holds 59 other files, including reference files (for example `README.md`, `agents/openai.yaml` and `examples/worked_examples.md`).

The repository describes itself as: This is for the userscripts created on GreasyFork.org. The licence is MIT.

When your agent uses it

  • Resolve an undecided choice
  • Narrow an exploration
  • Audit an invariant
  • Investigate an owner-identified gap

Example prompts

  • “/pick-invariant”

Workflow steps

6 steps, taken from the step headings in SKILL.md.

  1. Authority and mode
  2. Discovery, depth, and presentation
  3. Audit and structural commitment
  4. Hard specialist firewalls
  5. Commit and stop
  6. Progressive loading — activation semantics are part of the algorithm

What it can do on your machine

Read from SKILL.md and the folder at commit a6a319e. It shows what the files ask for, not the result of running them.

  • Tool permissions

    Pre-approves nothing: there is no allowed-tools line, so your agent's usual permission prompts apply.

    From allowed-tools in the SKILL.md frontmatter.

  • Runs code

    No scripts in the folder and no shell commands in SKILL.md.

    From the folder's file list and the shell code blocks in SKILL.md.

  • Network

    No URLs in SKILL.md.

    From URLs in SKILL.md, links to its own repository left out.

  • Credentials

    Names no API keys, tokens, secrets or passwords.

    From names ending in _API_KEY, _TOKEN, _SECRET, _KEY or _PASSWORD in SKILL.md.

Context cost

Pick Invariant loads about 2.2k tokens when it runs, and up to ~30k if it reads all its reference files. Until then it costs about 152 tokens; SKILL.md has 769 words of instructions outside code blocks.

Always · name and description, kept in context so the agent knows when to use it
~152
When it runs · the whole SKILL.md, loaded when a task matches
~2.2k
With references · SKILL.md plus every file in references/, read only if the agent opens them
~30k

Estimates: characters ÷ 4, the usual rule of thumb; real counts depend on the model's tokenizer. Scripts and assets cost tokens only if the agent reads them.

Safety

Auto-check passed

The automated check found no risky patterns in SKILL.md.

Automated static check — not a guarantee. Review scripts before installing. It scans the text of SKILL.md for risky patterns (piping downloads into a shell, reading credential files, hidden Unicode, destructive commands); files beside SKILL.md are not scanned.

SKILL.md

The full file from cyfung1031/userscript-supports at commit a6a319e, republished under its MIT licence (© cyfung1031). 769 words, ~2,191 tokens.

Download SKILL.mdSave it as .claude/skills/pick-invariant/SKILL.md (or your agent's skills folder). This skill also uses 55 other files; get the full folder from GitHub.
name
pick-invariant
description
Default first-line, domain-agnostic meta-reasoning control for choosing the right path. Use it to resolve an undecided choice, narrow an exploration, audit an invariant, investigate an owner-identified gap, or determine whether a validated owner can be used directly when no Pick exploration or audit is requested. Preserve target, authority, and semantics; retain only observable distinctions that can change the target decision; prefer the cheapest sufficient observation or representation; refine only on an explicit witness; and load specialist doctrine only when its trigger is material.

PickInvariant

Govern every path by:

Preserve target, authority, and semantics. Keep only decision-relevant distinctions. Prefer the cheapest sufficient observation/representation. Refine only on an explicit witness.

Pick remains the structural grammar:

text
R_P(x)=<P(x),I(x),B(x),χ(x)> ; D(x)=d(R_P(x))

P = applicability/authority/observable scope; I = owned local semantics; B = material seams; χ = genuinely non-local structure. Pick locates candidate distinctions; retention is target-relative:

text
x1 ~D x2 iff no admissible target-relevant discriminator requires different decisions
R(x1)=R(x2) => D(x1)=D(x2)

Use the coarsest evidence-supported observable representation. Retain a distinction only if removing or merging it collapses an acceptance-changing contrast or violates a bound requirement.

1. Authority and mode

Bind supplied artifact identity before exact-conformance claims; else mark INFERRED_PICKINVARIANT.

text
explicit authority > validated owner > validated extension > PickInvariant derivation > preference

Choose one primary mode:

  • BYPASS: adequate validated owner; no requested Pick exploration/audit. Run owner directly.
  • PICK_EXPLORE: explicit open exploration; non-authoritative. No audit pass, Δ, mutation, rebind, or authority installation.
  • PICK_AUDIT: explicit read-only stress test; preserve owner authority and bind audit depth.
  • PICK_DERIVE: only NO_OWNER or owner/higher-authority adopted PROCEDURE_GAP; bind INVESTIGATE | COMPILE. Only COMPILE may authorize action from a complete adopted gap.

Exploration/audit evidence never creates derivation authority.

2. Discovery, depth, and presentation

Exploration may reframe broadly while anchoring target, authority, observations, and constraints. Commit only target-bound observable contrasts, falsifiable questions, or adopted gaps.

The theorem stack is internal control logic, not mandatory user-facing ceremony. Unless architecture is requested or explanatory, answer in native domain language; do not dump P/I/B/χ, quotient classes, theorem names, certificates, or receipts merely to show work.

Depth:

  • DIRECT: validated owner/native mechanism; default for known structure.
  • FAST_DELTA: authorized derivation with one directly evidenced observable distinction, known decision effect, no role ambiguity, and no residual interaction; use owner + smallest Δ.
  • STRUCTURAL: role unknown, merge collapses a contrast, seams interact, information/history/time is nontrivial, or local facts + direct boundaries cannot reconstruct D.
  • COVERAGE: breadth obligation only for DELTA_AUDIT, FULL_AUDIT, or another explicit completeness claim.

Every escalation names its witness and new decision power; de-escalate once a simpler sufficient representation survives the relevant contrasts. Specialist probability/composition gates are orthogonal to depth unless they expose structural uncertainty.

3. Audit and structural commitment

For PICK_AUDIT, bind exactly FOCUSED_AUDIT | DELTA_AUDIT | FULL_AUDIT. Delta/full use SURVEY -> MAP -> CONTRAST -> PINPOINT until each material semantic family is CHECKED | EXCLUDED(reason) | UNVERIFIED(risk) or a named oracle blocks progress. Separate finding confidence from completeness; reconcile stale/resolved findings; never globalize scoped NO_GAP_FOUND.

Before structural derivation/rebinding bind domain, target, oracle, observables, exclusions, and Q_D; mark needed observables AVAILABLE_NOW | PROSPECTIVE_ONLY | ERASED_UNRECOVERABLE.

For each candidate distinction ask: admissible/reachable? target/consumer-changing? observable at decision time? does removal/merge collapse a valid contrast? Retain only if needed. Preserve seam coverage producer_guarantee >= consumer_requirement. Prove sufficiency before minimality; reserve MINIMAL for target-relative removal/merge-tested claims, else use SUFFICIENT or VALID_COMPRESSION.

When stochastic information-channel equivalence is material, use Blackwell/Le Cam rules in the probability/decision-quotient references; do not equate coordinate equality with information equality. When a global conclusion combines overlapping pieces, repeated decompositions, or non-local compatibility, activate COMPOSITION_TRIGGER; do not assume additivity. A validated owner that owns the law/composition stays DIRECT.

Show full SKILL.md (303 more words)Show less

4. Hard specialist firewalls

Activate PROBABILITY_MEASURE_TRIGGER only when probability law, sampling, conditioning/rejection, quotient weighting, stochastic transport, or information-channel semantics can change the target. Then load references/probability_semantics.md before any law-level, UNIFORM, stochastic- sufficiency, conditioning, transport, canonicality, or quotient-weighting claim. Incidental stochastic vocabulary does not fire it; a validated owner that already binds the material law may remain DIRECT.

Null/measure-zero or parameterization-sensitive conditioning requires an explicit conditioning / disintegration basis; symmetry-based canonicality requires the invariant-measure gate. If required detail cannot load, narrow/downgrade the claim rather than inventing semantics.

When several authorized observations could resolve uncertainty, prefer the cheapest reachable one that separates leading acceptance-changing alternatives; do not invent numeric priors/utilities. If a failed contrast may be abstraction-induced, validate it against raw/admissible state: real witness -> gap/failure evidence; spurious witness -> smallest witness-backed representation refinement. Never patch a known-insufficient representation only in the decision rule.

5. Commit and stop

Only PICK_DERIVE may run:

text
DELTA -> OBSERVE -> COMPILE -> FREEZE -> VERIFY -> REBIND -> STOP

Preserve mature-owner safeguards. Every nontrivial action must prospectively resolve/discriminate a named acceptance-changing condition. Executors consume compiled artifacts and do not rediscover PickInvariant. Rebind only on observed decision misclassification plus a named missing structural distinction. The certified gap is the derivation budget.

Use PICK_LITERAL only for an appropriate lattice polygon: A = I + B/2 - 1; otherwise PICK_STRUCTURAL. Never transfer literal coefficients without independent basis; Ehrhart/valuation extensions require their own assumptions.

6. Progressive loading — activation semantics are part of the algorithm

Load only on material triggers:

  • explore/rigor -> references/exploration_and_adaptive_rigor.md
  • Pick theorem/roles -> references/pick_abstraction.md, references/pick_representation.md
  • architecture/provenance -> references/theorem_provenance.md
  • merge/continuation/channel equivalence or MINIMAL -> references/decision_quotients.md
  • audit/coverage -> references/audit_and_contrast.md, references/review_scope_and_coverage.md; software DELTA/FULL -> references/software_audit_adapter.md
  • audit near-miss/residual -> references/audit_search_policy.md
  • canonicalization seam -> references/canonicalization_seam.md
  • partial commit -> references/partial_commit.md
  • interaction-heavy audit -> references/interaction_contrasts.md
  • derivation/failure -> references/procedure_gaps.md, references/binding_and_rebinding.md, references/execution_and_failures.md
  • architecture/resolution -> references/architecture.md, integration/procedure_resolution.md
  • seam/history/time -> references/seams_information_and_time.md
  • probability/measure -> references/probability_semantics.md
  • composition/observation-value/refinement -> references/composition_and_refinement.md
  • null-conditioning/symmetry/Ehrhart/local-global gate -> integration/specialized_gates.md
  • prompt/presentation -> references/prompt_and_presentation.md
  • robustness -> references/robustness.md
  • output/calibration -> templates/, examples/

Triggers are algorithmic. If required detail is unavailable, narrow or downgrade the conclusion rather than pretending it was applied.

© cyfung1031, MIT. Rendered from Markdown: HTML in the file is shown as text, images as links, and headings moved down two levels. Raw file

Files

SKILL.md and 55 other files (references) in agent-skills/pick-invariant of cyfung1031/userscript-supports.

  • SKILL.md
  • README.md
  • agents/openai.yaml
  • examples/worked_examples.md
  • integration/procedure_resolution.md
  • integration/specialized_gates.md
  • references/architecture.md
  • references/audit_and_contrast.md
  • references/audit_search_policy.md
  • references/binding_and_rebinding.md
  • references/canonicalization_seam.md
  • references/capability_library.md
  • references/composition_and_refinement.md
  • references/decision_quotients.md
  • references/execution_and_failures.md
  • references/exploration_and_adaptive_rigor.md
  • references/interaction_contrasts.md
  • … and 39 more

Open the folder on GitHubat commit a6a319e

Compare with similar skills

Pick Invariant next to the 5 skills that share the most tags, products or categories with it. Stars are the repository's; “used in” counts other GitHub owners with a copy.

Pick Invariant compared with similar skills
SkillStarsUsed inTokensAuto-checkLicenceRepo updated
Pick Invariant this skillcyfung1031/userscript-supports121—~2.2kAutomated safety check: PassMIT
Deepseek Reasonruvnet/ruflo74k—~626Automated safety check: NotesMIT
Sitemap Domainthedaviddias/Front-End-Checklist74k—~531Automated safety check: PassMIT
Domain Driven Designdavila7/claude-code-templates33k5 repos~623Automated safety check: PassMIT
Domain Modeling and Glossarywindmill-labs/windmill18k—~622Automated safety check: PassCustom licence
Flutter Cherry Pickflutter/flutter179k—~1.8kAutomated safety check: PassBSD-3-Clause

Similar skills

  • Deepseek Reason

    ruvnet/ruflo

    Reasoning-mode completion against DeepSeek's deepseek-reasoner model (R1) via /v1/chat/completions.

    74k GitHub stars~626 tokensUpdated today
    AI & LLM EngineeringAuto-check: notes
  • Sitemap Domain

    thedaviddias/Front-End-Checklist

    A skill your agent uses when applies to sites that have recently migrated from HTTP to HTTPS, changed domain name, or have www/non-www redirect configurations.

    74k GitHub stars~531 tokensUpdated 4 days ago
    Marketing & SEOAuto-check passed
  • Domain Driven Design

    davila7/claude-code-templates

    Plan and route Domain-Driven Design work from strategic modeling to tactical implementation and evented architecture patterns.

    33k GitHub starsUsed in 5 repos~623 tokens
    DevelopmentAuto-check passed
  • Domain Modeling and Glossary

    windmill-labs/windmill

    Actively challenges vague or conflicting terminology as you design, and keeps a living domain glossary file up to date in real time.

    18k GitHub stars~622 tokensUpdated yesterday
    DevelopmentAuto-check passed
  • Flutter Cherry Pick

    flutter/flutter

    How to land a formal cherry-pick of a merged PR for the flutter/flutter repo stable or beta channel.

    179k GitHub stars~1.8k tokensUpdated today
    MobileAuto-check passed
  • Invariant Guard

    sickn33/agentic-awesome-skills

    Correctness-first: forces writing the function contract, loop invariant, termination argument, and edge cases BEFORE code.

    47k GitHub starsUsed in 1 repo~4.5k tokens
    Auto-check passed

More from cyfung1031/userscript-supports

  • Review Userscript Change

    cyfung1031/userscript-supports

    Review browser userscript changes in JavaScript userscript files, focusing on scripting content and its HTML/CSS behavior, with metadata as a binding gate plus DOM/CSS, userscript-manager APIs…

    121 GitHub stars~1.8k tokensUpdated 2 days ago
    Auto-check passed
  • Prompt For Update 482487 Greasyfork Dark

    cyfung1031/userscript-supports

    Update the Greasy Fork Dark userscript's existing hard-coded // general CSS snapshot to the latest Greasy Fork application CSS while preserving the owner's dark palette, comments, selector-specific…

    121 GitHub stars~1k tokensUpdated 2 days ago
    Auto-check passed
  • Source Preserving CSS

    cyfung1031/userscript-supports

    Preserve-source CSS transformation for standalone CSS and CSS embedded in JavaScript, TypeScript, HTML, or user scripts.

    121 GitHub stars~764 tokensUpdated 2 days ago
    Auto-check passed
  • GitHub PR Publish Safe

    cyfung1031/userscript-supports

    Safe authenticated GitHub pull-request publication and review.

    121 GitHub stars~1.5k tokensUpdated 2 days ago
    Auto-check passed
  • Subagent Coordination

    cyfung1031/userscript-supports

    Coordinate bounded subagent work for analysis, coding, review, simulation, research, and handoff tasks.

    121 GitHub stars~1.2k tokensUpdated 2 days ago
    Auto-check passed

Questions about Pick Invariant

What does Pick Invariant do?

Default first-line, domain-agnostic meta-reasoning control for choosing the right path. Pick Invariant is an agent skill from cyfung1031/userscript-supports. Default first-line, domain-agnostic meta-reasoning control for choosing the right path.

When should I use Pick Invariant?

Pick Invariant fits situations like: resolve an undecided choice; narrow an exploration; audit an invariant; investigate an owner-identified gap.

How do I install Pick Invariant in Claude Code?

Run `npx skills add cyfung1031/userscript-supports --skill pick-invariant -a claude-code`. Or copy the skill folder (agent-skills/pick-invariant in cyfung1031/userscript-supports) into .claude/skills/pick-invariant in your project. Claude Code loads it when a task matches its description.

How do I install Pick Invariant in Codex?

Run `npx skills add cyfung1031/userscript-supports --skill pick-invariant -a codex`. Or copy the skill folder (agent-skills/pick-invariant in cyfung1031/userscript-supports) into .agents/skills/pick-invariant in your project. Codex loads it when a task matches its description.

Can I use Pick Invariant in Cursor, Gemini CLI or GitHub Copilot?

Cursor, Gemini CLI, GitHub Copilot and OpenCode also load SKILL.md folders. With the skills CLI, run `npx skills add cyfung1031/userscript-supports --skill pick-invariant -a cursor` (or -a gemini-cli, github-copilot or opencode for the others). To copy it by hand, put the folder in .cursor/skills/pick-invariant, .gemini/skills/pick-invariant, .github/skills/pick-invariant and .opencode/skills/pick-invariant in your project.

What does Pick Invariant need to run?

SKILL.md names no scripts, command-line tools or credentials: Pick Invariant is instructions for the agent only.

Does Pick Invariant access the network?

SKILL.md contains no URLs. Any network use would come from the scripts or tools the agent runs. This is read from the text; nothing was executed.

Is Pick Invariant safe to install?

Our automated static check of SKILL.md found no risky patterns, such as piping downloads into a shell, reading credential files or hidden Unicode. It is not a guarantee. Review the folder before installing.

What licence does Pick Invariant use?

Pick Invariant is published under the MIT licence (the repository's licence). It allows redistribution, so the full SKILL.md is shown on this page.

How many tokens does Pick Invariant use?

About 2.2k tokens (SKILL.md is roughly 8.8k characters). Agents keep only the skill's name and description in context until a task matches; then they load SKILL.md in full. Its references folder adds about 28k tokens, read only when the agent opens those files.

What are the alternatives to Pick Invariant?

Skills that share tags, products or a category with Pick Invariant: Deepseek Reason (ruvnet/ruflo, 74k stars), Sitemap Domain (thedaviddias/Front-End-Checklist, 74k stars), Domain Driven Design (davila7/claude-code-templates, 33k stars) and Domain Modeling and Glossary (windmill-labs/windmill, 18k stars). The comparison table on this page puts their stars, adoption, token cost, safety result and licence side by side.

Who maintains Pick Invariant?

cyfung1031 (a GitHub user) maintains it in cyfung1031/userscript-supports, which has 121 GitHub stars. The repository holds 6 skills in this directory. The repository was last updated on October 8, 2026.

Source: cyfung1031/userscript-supports on GitHub. Facts on this page come from the repository at the commit we read; the author's words are quoted as theirs.