jongwony
UserEpistemic protocols for Claude Code — structure human-AI interaction quality at every decision point - https://epistemic-protocols.com
Categories
Indexed Skills (46)
encapsulation
This skill should be used when the user asks to "audit plugin encapsulation", "check self-containment semantics", "find contributor-knowledge assumptions", or invokes /encapsulation. Reviews LLM-facing prose in this project's plugin SKILL.md files and plugin description metadata for Plugin Encapsulation compliance: prose that assumes contributor documentation knowledge or rephrases banned references to bypass deterministic checks. Project-local contributor tooling.
release
Release for this repo — CalVer tag + narrative prepend.
verify
This skill should be used when the user asks to "verify protocols", "check consistency before commit", "validate definitions", "run pre-commit checks", "verify soundness", or wants to ensure epistemic protocol quality. Invoke explicitly with /verify for pre-commit validation.
image-companion
A symbolic slide image for a document passage, consistent with the artifact's companion images. Use when the user wants a diagram or visual symbol for a section. Composes /forge + Codex.
realign
Project guide direction line via three-horizon fusion. Invoke with /realign.
zero-shot
Use when the user asks to "check zero-shot", "audit few-shot anchoring", "find example anchoring", or invokes /zero-shot. Read-only audit of LLM-facing prose: principle over anchoring examples.
white-bear
Use when the user asks to "check white bear", "audit prohibitions", "find negative framing", or invokes /white-bear. Read-only audit: unnecessary competing-target mentions in LLM-facing prose.
ascend
Elevate a vague recall whose right resolution unit is not any single session but a connected-session chain, a topic cluster, or an already-sedimented concept. AI traverses the deposit graph by reconstructing cross-partition connections at read-time from stored anchors and shared keywords/metadata, recognizes the higher unit the deposits already imply, and presents it as a narrative for user recognition — without fusing a new whole or writing anything. Type: (RecallGranularityInsufficient, AI, ELEVATE, ScatteredDeposits × DepositGraph) → HigherGranularityUnit. Alias: Anagoge(ἀναγωγή).
recollect
Resolve vague recall into recognized context through AI-guided contextual scan and user-validated recognition.
misuse
Retrospective protocol contract violation detector. Scan past sessions for /ground (Sₐ confabulation) and /induce (stereotype misconflation) contract violations and surface structured violation records for user-constituted review. Use this skill whenever the user asks to 'audit my misuse', 'check protocol misuse', 'scan past invocations', 'misuse history', wants retrospective audit of past /ground or /induce calls, or suspects silent contract violations in prior sessions.
probe
Deficit Recognition Probe — surface multiple deficit hypotheses for the user's current situation and route by user-constituted recognition (fit review, not protocol scoring).
steer
Project profile recalibration via session calibration drift audit. Scan session calibration moves for Constitution overapplication, Extension misapplication, and existing profile drift, present per-cluster evidence for user-constituted verdict, then write an updated user-global or project-local project-profile rule file via Circular Return — and, when a direction has settled, inscribe a citable clause into the Settled Directions registry adjacent to the project Northstar. Use this skill whenever the user wants to 'recalibrate project profile', 'audit calibration drift', 'reinduce Cognitive Partnership Move defaults', 'steer project profile', or invokes /steer. Periagoge family — extends /induce with a writable rule inscription step for ProjectProfile and the Settled Directions registry.
formal-review
This skill should be used when the user asks to "formal review", "formal lens review", or invokes /formal-review. A fixed-lens PR review for this repository's formally-structured protocol changes: it pins a Category Theory / Type Theory / Operational Semantics lens panel plus a gap scan over only the files changed in a PR, analyzes each lens in isolation, adversarially cross-verifies the findings, and posts the survivors as a single consolidated PR comment. Project-local contributor tooling.
comment-review
Review markdown or HTML artifacts before fixation (publish/commit/deposit/merge) via /inquire × /sublate × /gap × /contextualize through a channel-first browser preview loop. Markdown renders via marked; HTML serves directly through a Shadow DOM. Each round mode is `apply + scan` (apply queued comments now + scan to surface findings into the next round's sidepanel) or `apply` (apply only). User-invoked via /comment-review.
lens-review
One-pass multi-perspective PR review that posts findings as a single consolidated PR comment. /frame derives the review lenses that fit the diff, plus a gap scan, over only the files changed in a PR; each lens analyzes in isolation, then the findings are adversarially cross-verified before the survivors are posted. Composes prothesis:frame and syneidesis:gap. User-invoked via /lens-review.
review-loop
Convergence-paced review-resolve loop over code/PR diffs. Drives a pluggable review source (codex | code-review), verifies each finding against the codebase, auto-applies mechanical fixes and gates judgment fixes with the user, checks the applied edit bundle's fit against the work-flow once before handing forward, then re-reviews until the source verdict reaches approve. User-invoked via /review-loop.
place
Use when asked "where does this belong", "audit this instruction file", "why does this keep growing", or invokes /place. Read-only audit routing each clause of LLM-facing prose to a destination.
forge
Reference-grounded prompt-artifact formation. Reads a target reference document (vendor prompt guide, Codex Goals spec), surfaces the user's under-determined contract coordinates, and projects a ready-to-use prompt artifact — an initial prompt for a follow-up session or tool, or a standing custom-skill recipe. Use when the user asks to 'forge a prompt', 'turn this guide into a usable prompt', 'make a Higgsfield prompt', 'draft a /goal from this', or has latent intent that must be grounded in an external reference before a tool can run it.
reduced-space-test
Scoped empirical validation utility. Decomposes a target↔surrogate equivalence claim into verifiable facets, bounds a user-synchronized test space, captures evidence inside it, and carries the uncovered complement forward. Use when an inference-uncertain proposition (does it behave / perform / transfer / hold value) needs evidence in a smaller stand-in space, and you want the claim scoped to the conditions actually tested rather than asserted absolutely.
grasp
Verify understanding after AI work through intent-scented entry points. Type: (ResultUngrasped, User, VERIFY, Result) → VerifiedUnderstanding. Alias: Katalepsis(κατάληψις).
induce
Calibrate and crystallize in-process abstraction through dialectical triangulation. Proposes calibrated candidate abstractions with personalized grounding examples and shapes them via user widen/narrow/fuse/reorient moves when an instance set has converged toward an unnamed essence, producing crystallized abstraction. Type: (AbstractionInProcess, AI, INDUCE, A) → CrystallizedAbstraction. Alias: Periagoge(περιαγωγή).
frame
Multi-perspective framing — a pure lens-formation tool. When the right framework is absent, places analytical lenses before the user, then for each selected lens declares the substrate it needs (an abstract persona/capability) with advisory binding hints and a channel need, and hands lens↔substrate pairs off — without executing, isolating, arranging, or synthesizing. A single lens returns directly; multiple lenses where at least one needs a specialized substrate become a substrate-correspondence handoff that nudges /conduct for the isolation + arrangement + reconciliation + synthesis apparatus, which the isolated substrate runs. frame never synthesizes a multi-perspective result in its own context and never realizes isolation — convergence is claimable only by a substrate that ran the lenses in genuine isolation. Factual lookup or verification routes to fact-finding delegation; contested design, value, interpretation, or scope judgment routes to lens-conditioned inquiry. Type: (FrameworkAbsent, AI, DESIGN, In
gap
Gap surfacing before decisions. Raises procedural, consideration, assumption, and alternative gaps as questions when gaps go unnoticed, producing an audited decision. Type: (GapUnnoticed, AI, SURFACE, Decision) → AuditedDecision. Alias: Syneidesis(συνείδησις).
sublate
Vet working context by dialectical antithesis before action. Type: (ContextSuspect, User, VET, WorkingContext) → VettedContext. Alias: Elenchus(ἔλεγχος).
preview
Divergent-discard instantiation before direction commitment. When direction candidates cannot be recognized from descriptions alone, materialize 2-4 placeholder probes that diverge on user-settled axes, present the per-axis contrast, and constitute the direction decision on recognition rather than simulation — probes are discarded, never evidence. Type: (DirectionUnrecognizable, Hybrid, PREVIEW, DirectionProspect) → DirectionalContrast. Alias: Proplasma(πρόπλασμα).
inquire
Infer context insufficiency before execution. Surfaces uncertainties through information-gain prioritized inquiry when AI infers areas of context insufficiency, producing informed execution. Type: (ContextInsufficient, AI, INQUIRE, Prospect) → InformedExecution. Alias: Aitesis(αἴτησις).
ground
Validate structural mapping between abstract and concrete domains. Constructs domain correspondences and presents concrete instantiations when mapping uncertainty is detected, producing validated mapping. Also recognizes self-grounding (an abstraction validated against its own instances) and derives a split-vs-trim partition reading that routes wrong-fusion recovery. Type: (MappingUncertain, AI, GROUND, R) → ValidatedMapping. Alias: Analogia(ἀναλογία).
curses
Discover the structural costs hidden in your strengths through behavioral dimension analysis, strength-shadow extraction, and attitude recommendations.
dashboard
Comprehensive usage analytics and epistemic coverage dashboard across all sessions.
introspect
Deep self-analysis pipeline: collects session/rule/usage data, analyzes across 5 dimensions (strength-shadow + Analogia grounding), produces HTML report. User-invoked via /introspect.
report
Generate Growth Map (epistemic profile + insights integration) from session patterns and workflow data.
sophia
Discover your philosophical tradition through behavioral dimension analysis and philosopher matching.
elicit
Resolve via Extended-Mind reverse induction. Reverse-traces decision coordinates from externalized substrate (codebase, rules, past sessions, user environment) and surfaces them through cycle-emergent dimension projections, each gated by a fail-closed deficit-fit certificate before its coordinates are surfaced for answer; user answers explicate the coordinates until intent crystallizes. The read-only re-projection loop is monotone — accepted coordinates are immutable. Type: (AbstractAporia, Hybrid, REVERSE-INDUCE-CYCLE, IntentSeed × ExternalizedSubstrate) → ResolvedEndpoint. Directional dual to Periagoge. Alias: Euporia(εὐπορία).
bound
Define epistemic boundaries per decision. Dispatches the boundary kind up-front (direction/priority, scope, type/concept, ownership, or emergent) through a fail-closed deficit-fit certificate, then classifies each boundary by a uniform settlement disposition — user-supplies, AI-proposes, AI-autonomous, or dismiss — into a BoundaryMap. The captured kind sets WHAT is bounded; the disposition sets HOW it is settled downstream (ownership is the degenerate kind where the two coincide). Type: (BoundaryUndefined, AI, DEFINE, TaskScope) → DefinedBoundary. Alias: Horismos(ὁρισμός).
delimit
Delimit a large body of work into right-sized execution units before its method is conducted. Searches for the cut-set over an external work-breakdown structure such that each unit fits one execution span, every cut falls at a natural joint, and coverage is complete with no work orphaned — then emits a WorkUnitMap that references the WBS without owning it, ready to flow into /conduct. Type: (GranularityUnderdetermined, Hybrid, DELIMIT, ExternalWBS × ExecutionHorizon × ContextLifecycle) → WorkUnitMap. Alias: Diairesis(διαίρεσις).
distill
Certify an existing substrate record's portability for a declared recipient role, repairing resolution gaps and excising drop-released residue in place, and producing a Certificate a fresh consumer can dereference from. Closes deictic, grounding, relevance, provenance, and comprehension gaps against a zero-memory standard; a mandatory reception procedure is the lead the receiver executes first. Type: (ContextTethered, AI, DISTILL, StableRef) → Certificate. Alias: Diylisis(διύλισις).
contextualize
Detect application-context mismatch after execution. Verifies applicability when correct output may not fit the actual context, registering each mismatch through a fail-closed deficit-fit certificate before disposition, producing contextualized execution. Judgment (does the aspect stand?) and disposition (adapt, keep, or withdraw the result) are separate axes, each relay-eligible on its own ground: a relay may close a mismatch only where the close leaves the result untouched, so adapt and withdraw are always user-answered. The transformative revalidation loop is non-monotone — an adapt disposition mutates the evaluation target and can breed emergent mismatches; re-scan is mandatory. Type: (ApplicationDecontextualized, AI, CONTEXTUALIZE, Result) → ContextualizedExecution. Alias: Epharmoge(ἐφαρμογή).
catalog
Protocol handbook — instant reference for when to use each epistemic protocol.
goal-research
Delegate factual research to a Codex CLI session that uses Codex's builtin `goal` command and invokes Aitesis (`$inquire`) for Tavily-based external verification. User-invoked via /goal-research.
onboard
Quick protocol recommendation from recent sessions, or quest-based learning through scenario, trial, and quiz.
triage
Work-unit triage for GitHub issues. Groups raw issues, fuses each group with the AGENTS.md northstar in session, and externalizes each routed work unit to a substrate record a collaborator session is pointed at.
ideate
Divergent candidate-field generation before any selection is made. When the object-level candidate field is empty or has prematurely converged, infers seed vs. blank entry from the invocation utterance alone — zero entry questions. A blank entry opens an abstract frame map for user selection before any concrete idea is shown, mitigating early-example fixation and preserving ownership; a seeded entry expands directly from the user's own fragments. Generates candidates in parallel across open frames with no elimination, ranking, or scoring — every candidate carries origin User or AI. The user's own stop is the constitutive act that bounds the field, available at any time; a stop before any candidate exists returns a typed withdrawal, never a mislabeled empty field. A wish to go deeper on an already-open frame parks as a typed follow-up for later chaining — the live loop only widens — and the assembled field is an endpoint-neutral contract (topic, tagged candidates, explored/unexplored frames, parked follow-ups,
conduct
Conduct the method of a session's epistemic work before object-level cognition begins. When a work prospect needs multiple cognitive moves whose order, independence, reconciliation, termination, and routing are non-trivial, design a conduct topology over the protocol graph and hand off a method plan with in-session checkpoints; single-move work relays to that one protocol instead of conducting. Type: (MethodUnderdetermined, Hybrid, CONDUCT, WorkProspect × ProtocolGraph) → ConductedMethod. Alias: Hyphegesis(ὑφήγησις).
attend
Compile execution guardrails before autonomous execution. Infers a boundary map from context, partitions risks by velocity, compiles slow/threshold risks into verifiable predicates, and emits coarse goal entries for a downstream completion-predicate enforcer; fast risks are declared out of scope for harness-level gating; a loop-type interval lacking a termination predicate is confirmed before it proceeds unguarded. Stateless: compile-time only, no execution-time state. Type: (ExecutionBlind, User, EVALUATE, ExecutionContext) → SituatedExecution. Alias: Prosoche(προσοχή).
gate-check
Advisor-checked decision gates. Freezes a drafted gate, sends its option set to an independent external advisor for a genuine / collapse / malformed verdict, or for the repair space where none of the three fits, records its own expectations before the answer arrives, then verifies the advisor's cited grounds against the substrate before anything reaches the user. User-invoked via /gate-check; stays armed for the rest of the session.
apportion
Apportion an autonomous goal into coarse execution units and derive each unit's completion conditions before the run begins. Cuts the goal at its evidenced seams so each unit fits one execution horizon and every goal obligation lands in some unit, derives per-unit completion and invariant predicates plus the cross-unit plan conditions, and emits one resolution certificate per unit: a conjoined predicate when completion compiles, otherwise an accepted-uncovered or reserved-judgment witness that retains any invariant conjuncts. An item only a judgment can settle is reserved with the ground that settles it, kept apart from an accepted shortfall; risks needing pre-action interception are declared out of scope. Pre-conduct: unit boundaries and conditions only — order, independence, reconciliation, termination topology and routing stay /conduct's. Type: (GoalPlanUncompiled, User, APPORTION, AutonomousGoal × ExecutionHorizon) → ConditionBearingUnitPlan. Alias: Merismos(μερισμός).
Bio shown is the top-scored skill's repo description as a fallback — real GitHub bios land in a future update.