Thirsty Skill Library

A governed, composable library of 79 specialist skills for Thirstys Workshop. Every skill shares one constitutional baseline — provenance, uncertainty, reversibility, approval, and handoff rules — and ships as a self-contained skill.zip for ChatGPT Skills.

Skills are progressively being assessed and reconstructed under the Thirstys Skill Standard; each skill's maturity, class, version, and evaluation status are recorded in CATALOG.json. An UNASSESSED maturity means the skill predates the standard and has not yet been assessed against it — not that it is unusable.

How it works

Skill sources are small and declarative. The two constitutional reference files live once in shared/ and are injected into every package at build time, so amending the constitution is a one-file change that reaches all 79 skills on the next build.

flowchart LR
    A["skills/&lt;name&gt;/<br/>SKILL.md + agents/"] --> B["build.ps1"]
    S["shared/references/<br/>constitutional-rules.md<br/>handoff-schema.md"] --> B
    B --> Z["dist/&lt;name&gt;/skill.zip"]
    Z --> U["ChatGPT Skills upload"]

Loading

Quick start

./build.ps1                     # package all skills into dist/<name>/skill.zip
./build.ps1 -Skill <name>       # rebuild one skill
./build.ps1 -Clean              # wipe dist/ and rebuild everything
./validate.ps1                  # check structure, frontmatter, catalog, and docs sync
./new-skill.ps1 -Name <name> -Description ... -Objective ... -Outputs ...
                                # scaffold a new skill and register it in CATALOG.json

Install by uploading one dist/<name>/skill.zip at a time in ChatGPT Skills.

Repository layout

Path	Purpose
skills/<name>/SKILL.md	The skill itself: frontmatter (name, description) plus its objective, method, output contract, quality gate, and composition guidance
skills/<name>/agents/openai.yaml	Display metadata for the ChatGPT Skills interface
skills/<name>/references/, assets/, scripts/	Optional package depth: domain references, output schemas/templates, deterministic scripts (packaged into the zip automatically)
shared/references/	Single source of truth for the constitutional rules and the inter-skill handoff schema
CATALOG.json	Machine-readable index: identity, objective, outputs, group, class, version, maturity, and validation record for every skill
library-schema.json	Canonical enums and thresholds (maturity, classes, groups, outcomes, eval minimums) consumed by all three scripts
docs/	The Skill Standard, class taxonomy, depth scorecard, capability audit, and composition recipes
evals/	Per-skill evaluation suites: trigger/functional/boundary cases, fixtures, and committed result records (how they work)
dist/	Build output (gitignored) — regenerate with ./build.ps1; each zip ships with a skill.zip.sha256 package-identity checksum
.github/workflows/ci.yml	CI: runs validate.ps1, check-evals.ps1 (structural only), and build.ps1 on every push and PR, publishing the zips as a build artifact

Constitutional baseline

Every skill operates under the same operating rules (shared/references/constitutional-rules.md): distinguish fact from inference, preserve provenance, never invent missing context, separate analysis from consequential action, prefer reversible moves, and end with residual uncertainty and the next highest-value action. Structured handoffs between skills follow shared/references/handoff-schema.md.

A skill may override a shared file by shipping its own copy at the same relative path — validate.ps1 flags every override so none go unnoticed.

Library standard

The Thirstys Skill Standard defines what a finished skill package contains: a capability delta (what the skill changes versus unaided execution), explicit trigger and non-trigger conditions, an input contract with minimum evidence, a specialist method, a failure taxonomy, negative space (when the skill returns a no-result outcome instead of forcing a match), an output contract, and class-appropriate evaluations. Maturity (UNASSESSED → DRAFT → EXPERIMENTAL → REVIEWED → VALIDATED → STABLE) is a summary of a per-skill validation record — structural checks, executed trigger/functional/boundary evaluations, human review, and real-use evidence are tracked as separate dimensions, and maturity may never claim more than the record shows. STABLE requires evidence of actual use; certification here means exactly structural validation plus recorded evaluations plus review — nothing cryptographic or formally verified is implied.

Supporting documents: the class taxonomy, the depth scorecard, the capability audit, composition recipes, and the Wave 1 evidence report. Worked examples show what the first-wave skills actually produce.

Skill catalog

Orchestration & routing

Skill	Use it to
thirsty-work-orchestrator	Coordinate end-to-end Thirstys Workshop workflows across context routing, ingestion, synthesis, governance, engineering, research, publication, and hostile review.
thirsty-context-router	Resolve ambiguous project terms, select the correct context, and route requests to the right specialist skill — or return an explicit no-match outcome (AMBIGUOUS, UNSUPPORTED, INSUFFICIENT_EVIDENCE, …) instead of forcing a fit.
thirsty-workspace-ingest	Inventory and classify uploaded folders, repositories, papers, notes, and archives. Use when the user asks to ingest, map, catalog, normalize, or prepare a workspace for downstream analysis.
thirsty-knowledge-synthesizer	Synthesize grounded knowledge across repositories, papers, specifications, websites, and internal documents. Use when asked for a state-of-knowledge brief, concept trace, cross-source synthesis, or gap analysis.
q-orchestrator	Coordinate Q hostile review using assumptions, epistemic status, contradictions, absent stakeholders, power, falsification, reversibility, mission drift, and attention allocation. Use when the user addresses Q or asks what they are not seeing.

Constitutional design & testing

Skill	Use it to
constitutional-system-architect	Design or evaluate governed-AI constitutional systems. Use for constitutions, authority models, rights, duties, amendment processes, enforcement, appeals, and emergency powers.
constitutional-red-team	Perform hostile review of constitutional and governance structures. Use when asked to attack, red-team, stress-test, capture-test, or identify abuse paths in governed systems.
constitutional-test-generator	Generate tests from constitutional language and governance requirements. Use for compliance, adversarial, boundary, delegation, memory, emergency, amendment, and recovery scenarios.
constitutional-diff-interpreter	Interpret the governance meaning of changes between constitutional versions. Use when reviewing amendments or comparing versions.
amendment-impact-simulator	Evaluate proposed constitutional amendments before adoption. Use to model direct, second-order, implementation, migration, and attack-surface effects.
constitutional-precedent-manager	Organize and reason over prior governance decisions as precedent. Use to find relevant cases, compare rulings, and detect inconsistent treatment.
constitutional-translation-layer	Translate between constitutional provisions, legal language, software policy, agent instructions, runtime controls, and human-readable rights.
scenario-constitution-tester	Stress-test constitutions through extended social and institutional scenarios such as scarcity, conflict, capability shifts, misinformation, succession, and collapse.
counterfactual-governance-simulator	Compare how alternative governance structures might have handled an event. Use to test centralized, federated, delegated, or rights-based alternatives.
incident-constitutional-review	Run a constitutional postmortem after failure or near miss. Use to distinguish constitutional, interpretation, enforcement, evidence, authority, and oversight failures.

Governance auditing

Skill	Use it to
governance-traceability-auditor	Audit traceability from governance principles through requirements, policies, controls, code, tests, and evidence. Use when verifying whether implementation supports stated governance claims.
jurisdiction-boundary-mapper	Map where human, agent, institutional, and constitutional authority begins and ends. Use to find overlaps, gaps, circular escalation, and undefined sovereignty.
delegation-chain-auditor	Trace delegated authority through principals, agents, subagents, tools, and external systems. Use to verify scope, duration, revocability, and accountability.
rights-remedy-auditor	Check whether every stated right has detection, standing, appeal, enforcement, restoration, and remedy. Use to identify merely aspirational rights.
emergency-powers-auditor	Audit exception and emergency powers. Use to inspect triggers, declaring authority, scope, duration, oversight, renewal, sunset, and restoration.
legitimacy-modeler	Evaluate why affected parties should regard a governance system as legitimate. Use for consent, participation, competence, procedure, outcomes, representation, and rights protection.
institutional-capture-simulator	Model gradual capture of governance institutions. Use for appointment, metric, dependency, funding, data, emergency, expertise, and procedural capture scenarios.
governance-debt-register	Track accumulated governance compromises such as unenforced policy, stale grants, missing appeals, temporary exceptions, and untested clauses.
power-distribution-analyzer	Map actual power rather than formal authority. Use to analyze control over compute, data, deployment, funding, credentials, audits, appeals, communication, and shutdown.
incentive-compatibility-reviewer	Assess whether actors benefit from following governance rules or are rewarded for evasion.
metric-corruption-auditor	Red-team metrics, benchmarks, KPIs, and thresholds for gaming, proxy failure, invisibility, and goal distortion.
dependency-sovereignty-auditor	Find external dependencies that can undermine independence or constitutional guarantees, including providers, hosting, identity, registries, data, funding, domains, and trust anchors.

Agent & AI governance

Skill	Use it to
governed-agent-designer	Design agents with explicit purpose, jurisdiction, authority, tools, memory, delegation, escalation, refusal, audit, shutdown, and recovery boundaries.
human-agi-relations-reviewer	Evaluate systems for durable human-AGI relations. Use for agency, consent, reciprocal duties, power asymmetry, representation, exit rights, conflict resolution, and moral uncertainty.
capability-change-impact-review	Reassess governance when an AI system gains materially new autonomy, planning, persuasion, tool use, memory, replication, or social influence.
identity-continuity-auditor	Assess agent identity continuity across model upgrades, forks, restores, migrations, and memory changes.
memory-governance-designer	Design constitutional rules for agent memory admission, provenance, correction, expiry, access, isolation, inheritance, deletion, and contested records.
model-agent-provenance-auditor	Track what model, instructions, tools, memory, policies, data, and human interventions produced an output.
multi-agent-treaty-designer	Design agreements between independently governed agents or systems. Use for recognition, jurisdiction, information exchange, noninterference, verification, breach, arbitration, and withdrawal.
protocol-interoperability-auditor	Audit whether separately governed systems can interact safely. Use to find incompatible authority, identity, consent, evidence, and enforcement assumptions.
consent-revocation-designer	Design and audit meaningful consent and withdrawal mechanisms. Use for informed, specific, voluntary, comprehensible, revocable, time-bounded, enforceable consent.
decision-reconstruction	Reconstruct consequential decisions from available evidence. Use to explain what information, rules, interpretations, tools, and interventions shaped an outcome.

Epistemics & decision discipline

Skill	Use it to
assumption-registry	Create and maintain a live inventory of project assumptions. Use to register, review, expire, challenge, or trace dependencies on assumptions.
epistemic-status-enforcer	Apply explicit epistemic status to meaningful claims. Use to separate observed, verified, reported, inferred, hypothesized, speculative, normative, contested, and unknown statements.
contradiction-engine	Find incompatible statements across repositories, papers, policies, conversations, and specifications. Use for contradiction, drift, and terminology conflict analysis.
concept-lineage-tracer	Trace how an idea, term, mechanism, or policy evolved across time and artifacts. Use to locate origins, revisions, renamings, forks, and current authority.
semantic-canonicalizer	Maintain canonical terminology across projects. Use to build glossaries, normalize terms, identify overloaded language, and deprecate ambiguous synonyms.
moral-uncertainty-register	Track unresolved moral assumptions and how decisions change under alternative ethical frameworks.
value-conflict-adjudicator	Structure conflicts between values such as autonomy, safety, privacy, fairness, and collective welfare.
stakeholder-absence-detector	Identify affected but absent or unrepresented stakeholders, including future persons, nonusers, maintainers, dependent institutions, and artificial agents.
hostile-unknowns-review	Identify what is omitted, unasked, assumed, strategically ignored, or rendered invisible in a plan, architecture, paper, or decision. Use when the user asks Q, what am I not seeing, hostile review, or challenge the framing.
uncertainty-triggered-escalation	Define when missing information, conflicting evidence, model uncertainty, normative ambiguity, unknown authority, or irreversible consequences require escalation.
decision-threshold-controller	Define evidence and approval thresholds for reversible experiments, public claims, releases, amendments, partnerships, and irreversible deployments.
reversibility-exit-review	Evaluate whether actions can be undone and whether affected parties can meaningfully exit.

Research & publication

Skill	Use it to
research-claim-auditor	Audit scholarly and technical claims for evidence, overstatement, falsifiability, citation gaps, and reproducibility. Use before submission or publication.
research-publication-pipeline	Coordinate manuscript and release preparation across papers, repositories, metadata, DOI, ORCID, SSRN, Zenodo, and project sites. Use when preparing work for publication.
prior-art-literature-mapper	Map current literature, prior art, and adjacent work around a concept or mechanism. Use for novelty assessment, research positioning, and terminology discovery.
paper-repository-consistency-checker	Compare papers, repositories, diagrams, specifications, release notes, and public descriptions for consistency. Use to detect stale claims, terminology, or implementation drift.
research-question-generator	Generate research questions from contradictions, gaps, failed tests, assumptions, and decision needs.
experiment-governance-designer	Design experiments with explicit technical and constitutional boundaries, evidence thresholds, monitoring, rollback, and expansion criteria.
evidence-portfolio-manager	Track positive, negative, contradictory, missing, and replicated evidence supporting major claims.
falsification-planner	Design the strongest practical attempts to prove a theory, architecture, or governance claim wrong.
novelty-boundary-auditor	Separate genuine novelty from recombination, renaming, parallel discovery, or overlooked prior art.
proof-obligation-generator	Identify what must be demonstrated before accepting architectural, safety, governance, or research claims.
public-claim-risk-reviewer	Review public statements for unsupported certainty, accidental commitments, terminology risk, misinterpretation, and inconsistency.
critic-response-builder	Process serious criticism by steelmanning objections, separating valid points from misunderstandings, and planning corrections or response.

Engineering & architecture

Skill	Use it to
repository-architecture-cartographer	Map a repository as an architectural system. Use when asked to explain components, dependencies, trust boundaries, data flows, control flows, or governance enforcement locations.
spec-to-implementation-planner	Convert research, governance, or product specifications into executable engineering plans. Use for epics, issues, interfaces, schemas, acceptance criteria, and phased implementation.
architecture-decision-recorder	Turn discussions, experiments, and changes into durable architecture decision records. Use when recording why a choice was made and what would reverse it.

Strategy, partnerships & outreach

Skill	Use it to
mission-drift-detector	Compare current projects, priorities, partnerships, and behavior against declared mission and governing commitments.
strategic-coherence-auditor	Check whether research, repositories, publications, partnerships, and product work reinforce one another.
attention-allocation-advisor	Prioritize limited attention using uncertainty reduction, constitutional importance, dependency centrality, leverage, urgency, and opportunity cost.
project-fragmentation-detector	Detect when one conceptual project has split into incompatible repositories, branches, specifications, or communities.
unfinished-work-recovery	Find abandoned but potentially valuable work across branches, drafts, issues, notes, and publications.
partnership-due-diligence	Grade a prospective collaborator, platform, or funder across six diligence dimensions on ranked dated evidence — UNKNOWN scored as risk, exit cost modeled, posture set by decision table.
stakeholder-translation-engine	Translate technical and governance work for researchers, architects, policymakers, partners, or the public without weakening meaning.

Library meta-governance

Skill	Use it to
thirstys-skill-developer	Take a capability from intent to a validated, well-triggered skill package that meets the Skill Standard — or convert a procedure, standard, or checklist into one. Owns the whole authoring lifecycle as a single job.
skill-portfolio-auditor	Audit the library as a whole: coverage gaps, overlap and trigger collisions, specialization classification via the blind test, and per-skill add/deepen/merge/split/deprecate dispositions.
skill-constitution-auditor	Audit whether a skill complies with governing principles for authority, tools, data, provenance, uncertainty, escalation, and user agency.
skill-composition-auditor	Audit whether combining individually acceptable skills creates contradictory, unsafe, circular, or overly powerful workflows.
skill-permission-manifest	Generate and validate machine-readable declarations of skill connectors, reads, writes, external effects, approvals, retention assumptions, and allowed calls.
skill-observability-ledger	Define or review records of significant skill invocations, sources, tools, decisions, outputs, effects, uncertainty, and governing versions.
skill-regression-tester	Run a skill's evaluation suite with the class-appropriate method, compare to baseline and prior version, and record provenance-complete results — the eval evidence maturity depends on.
skill-drift-detector	Detect when real usage exceeds a skill intended scope or users rely on unintended behavior.
skill-dependency-resolver	Map skill dependencies, call permissions, cycles, foundational services, and version constraints.
skill-evolution-manager	Decide a skill's version bump, evidence-backed maturity transition, and release readiness, and manage deprecation with replacement pointers.

Contributing

See CONTRIBUTING.md. The short version: ./new-skill.ps1 to scaffold, edit the generated SKILL.md, add the skill to the catalog table above, then ./validate.ps1 and ./build.ps1 must both pass.