agent skill registry

Skill definitions, served raw for agents.

raw SKILL.md text/markdown · CORS-open manifest signed

index.json manifest signature llms.txt TGP.SKI

what this is

Each skill's full directory — SKILL.md, references, scripts, assets — mirrored from my public repositories at stable URLs. The manifest carries the source repo, the pinned commit, and a sha256 per file, and is signed with the key published on TGP.SKI.

use

# the manifest lists every skill: url, source repo, commit, sha256
curl -s https://skills.TGP.SKI/index.json

# verify the manifest against the key published on TGP.SKI
curl -sO https://skills.TGP.SKI/index.json
curl -sO https://skills.TGP.SKI/index.json.asc
curl -s https://TGP.SKI/keys/tgpski.asc | gpg --import
gpg --verify index.json.asc index.json

# fetch any skill raw (text/markdown, CORS-open)
curl -s https://skills.TGP.SKI/skills/leather/release-prep/SKILL.md

Skill files are verbatim mirrors — the sha256 in the manifest matches both the served file and the blob at the pinned source commit on GitHub.

the registry

TGPSKI/abductive-triage

abductive-triageStructured diagnostic reasoning for AI agents investigating incidents, failures, and unexpected behavior.762e2accab46

TGPSKI/adherence-suite

adherence-triageDiagnose a failing or suspicious adherence-suite result: is it the model, the grader, the adapter, the proxy, or the coordinate system?5ba2547faa91

onboard-fixtureOnboard a real repository as an eval fixture and build an eval on it: screen, vendor, prove offline, define arms and controls, freeze ground truth, pilot.5ba2547faa91

TGPSKI/catnip

catnip-onboardingConfigure catnip for a new user, from nothing installed to collecting on a schedule.f10304fde4f2

catnip-prowlHunt through catnip's collected traffic data for what actually happened — trends, standout anomalies and their causes, coupled repos, audience versus fetchers, and the things strict arithmetic gates out.f10304fde4f2

catnip-triageDiagnose catnip problems — empty charts, missing repos, a timer that stopped, data that looks wrong, onboarding that will not complete.f10304fde4f2

TGPSKI/claim-analysis

claim-analysisAnalyze claims in any conversational record (chat thread, email chain, meeting notes, transcript, review comments) for factual accuracy and distortion patterns.0b1f10ff854c

TGPSKI/conceit

build-triageInvestigate and prevent build failures for conceit from-source builds (PyTorch, vllm, llama.cpp).e3ac58892961

env-intakeWalk a user through conceit's environment intake on a new machine: discover every CUDA toolkit and host compiler present, resolve which pair nvcc will actually accept, prove it with a compile, and persist the answers to conceit.env.e3ac58892961

TGPSKI/directed-contexts

directed-contextsGenerate bounded codebase ownership contexts that agents can load inline or spawn in isolation.fa3d912a1d6a

TGPSKI/directed-workflows

directed-workflowsCreates directed workflow files for any repository.79a64a03a5b8

TGPSKI/dither

splash-ascii-renderConvert a reference image into a terminal splash screen (render/<name>/show.sh) rendered as truecolor Unicode quadrant-block art, sized to the caller's terminal.3794307beb63

TGPSKI/friction-report

friction-reportDetect and score friction in a conversational record using an evidence hierarchy and anti-pattern catalog.f58faac24f57

TGPSKI/go-fips-toolkit

fips-crypto-auditRuns a FIPS 140-3 crypto compliance audit across Go repositories, classifying every crypto/ import by risk and generating per-component reports.99a69738bd17

fips-finding-triageClassifies an individual FIPS crypto audit finding as real cryptography or a non-cryptographic use of a crypto primitive, using structured diagnostic reasoning.99a69738bd17

fips-remediation-planGenerates component-specific FIPS remediation TODO lists from crypto audit results, with effort estimates and suggested PR sequencing.99a69738bd17

TGPSKI/last-call

last-callAudit a private or local repository for everything that becomes permanent when it goes public: personal and machine leaks, references to files nobody outside can read, documentation that has drifted from the shipped code, features that shipped undocumented, a security surface described smaller than it is, and a commit history whose subjects and bodies expose the author's working process.592f2e96d5a5

TGPSKI/leather

agents-doc-lifecycleManage AGENTS.md and .subagents/*.md across the leather subagent guide set (currently 17 guides spanning core, runtime, worker, serve, tannery, quality, performance, security, operations, shell-mcp, ui, replay, agent-definition, tools/skills/toolsets semantics, integrations, examples, and observability).47221bdfb4cd

code-quality-lifecycleAudit and improve test coverage, formatting, linting, and test organization for the leather codebase.47221bdfb4cd

documentation-lifecycleGenerate, update, and maintain per-module documentation under docs/modules/, keep docs/ARCHITECTURE.md and README.md current, and produce Mermaid diagrams for data-flow, block, and dependency views.47221bdfb4cd

release-prepPrepare a leather release: auto-detect next version from git history, insert CHANGELOG section, update docs, commit and push.47221bdfb4cd

release-tagTag a prepared leather release and push the tag to origin.47221bdfb4cd

TGPSKI/olwb-micro-plugin

olwb-benefitsCreate, configure, and troubleshoot olwb's benefits features: destinations (/dest), sending (/send), and the notes → agent-work issues pipeline (/issues).a5c342b55078

TGPSKI/record

recordRecord terminal demos of CLIs and TUIs and publish them as README-ready GIFs — interactively via asciinema/agg, or fully unattended by driving the recorder inside a pty with a timed key script.bda647377597

TGPSKI/redline

redlineFinds the maximum safe --max-model-len and --max-num-batched-tokens for a vLLM-served model on a fixed GPU, by climbing a doubling ladder one clean reboot at a time and recording the real KV-cache/concurrency signal from each boot rather than guessing.673957879d34

TGPSKI/run-watcher

run-watcherBuild a live-render TUI that is the control center, debugging surface, and anomaly detector — for a long-running job, or for a continuous system like logs or traffic where the question is not 'did it work' but 'is this normal'.e37f632d3c67

TGPSKI/security-context-spec

generate-security-contextGenerate or update a .security-context.yaml file for any repository.d22e2a9dae59

TGPSKI/skeptic

cli-flag-reviewAudit and optimize CLI flag definitions across all skeptic subcommands.a663db054a63

doc-accuracy-auditAudit a document against the codebase for accuracy, then tighten it: verify claims, cut drift, simplify, restructure for the reader.a663db054a63

go-releaseManage the full Go module release lifecycle: determine next semver version, update CHANGELOG.md, create and push a git tag, and verify pkg.go.dev picks up the new version.a663db054a63

ruleset-lifecycle-operationsAdd or update rule definitions, create and manage rule groups, and migrate rules across groups in skeptic using the property-based layout.a663db054a63

threat-rules-ingestionIngest fresh security research from URLs/files, generate signed scanner rule packs, and validate with policy and behavior scan styles across ecosystems.a663db054a63

workflow-action-updatesAudit and update GitHub Actions workflow files to use the latest action releases pinned to full commit SHAs.a663db054a63

TGPSKI/words-are-cheap

words-are-cheapRead a named body of words — a commit range, a document, a package's comments, a set of error strings, help text, a PR description — and report which lines inform and which perform.da0b73d4cce9