Christian-Katzmann/pdoom-skills

Perfil de Christian-Katzmann

Última atividade de origem registrada: 8 de setembro de 2026 às 17:20

Catálogo do SkillsMP atualizado: 17 de outubro de 2026 às 00:09

skills coletadas: 44

Estrelas no GitHub: 0
Forks no GitHub: 0

README do repositório

Arquivo de origem do README: README.md

Ler o README original

Expandir README completo

# (p)doom-skills 18 methods and local tools for Codex and Claude Code, selected and adapted by Christian Katzmann to help agents investigate, decide, build and verify. **Status:** Runnable examples and recorded checks are included for all 18 packages. Small model comparisons include ties, mixed results and failed attempts; they do not establish a general success rate. [Browse the website](https://christian-katzmann.github.io/pdoom-skills/) · [Find a skill](#find-a-useful-skill) · [Install one](#install-one-skill) [![Product Walkthrough: the sample app confirms a saved note, but after reload the note field is empty.](design/screenshots/walkthrough-result.png)](packages/product-walkthrough/examples/walkthrough.md) *Product Walkthrough caught a deliberately planted save failure in Pocket List. These are captures from the actual browser run, with synthetic data and partial coverage. [Replay the run](packages/product-walkthrough/examples/walkthrough.md) or [regenerate this image](design/screenshots/render.mjs).* Each skill supplies instructions and, where useful, scripts or references for your coding agent. Read its example and limitations, then install only what fits your work. The website lets you inspect the collection; skills run through your local agent and any tools or services they require. ## Install one skill With Git and Python 3.12 available: ```sh git clone https://github.com/Christian-Katzmann/pdoom-skills.git cd pdoom-skills python3 -m venv .venv .venv/bin/python -m pip install -r tools/requirements.txt ./install list ./install install thinking-hats --host codex --dry-run ./install install thinking-hats --host codex ``` For Claude, replace `--host codex` with `--host claude`. The installer displays dependency requirements, refuses unrelated or edited files, and retains reversible transactions. Use `--dest` to choose the exact skills directory. On macOS, temporary destinations must use resolved paths such as `/private/tmp/`, because `/tmp/` is a symlink. See [installation and reversal](tools/README.md). In Codex, invoke `$thinking-hats` with your actual decision and relevant evidence. For example: `$thinking-hats Review whether this import can safely be retried. Inspect the importer and its tests, identify duplicate-write risks, and propose the smallest useful check.` In Claude, use `/thinking-hats`. Each [package README](packages/thinking-hats/README.md) lists its exact entrypoints, prerequisites and first use. Installation tests verify directory layouts; they do not prove discovery or reasoning quality in every client version. ## Find a useful skill ### Thinking and decisions | Skill | Use it for | | --- | --- | | [Bottleneck](packages/bottleneck/README.md) | Find the limiting business step or the next measurement needed to choose it, without making guesses into proven constraints. | | [Expert Mode](packages/expert-mode/README.md) | Finds the specific specialty a question belongs to, researches only when a named knowledge gap could change the answer, and takes a position with its trade-off. | | [Force Multiplier](packages/force-multiplier/README.md) | Finds a specific change that makes existing assets more useful, with observed drag, an amplification mechanism, the first move and work to stop. | | [Next Moves](packages/next-moves/README.md) | Compares what a product promises with what its repository contains, then names the few capabilities its users would most obviously reach for and not find. | | [Thinking Hats](packages/thinking-hats/README.md) | Chooses one focused reasoning lens before coding, so the lens decides what the agent inspects first, distrusts, writes before coding and verifies. | | [Vision](packages/vision/README.md) | Recovers what a project is actually about from its own history, tells a chosen change of direction apart from unnoticed drift, and tests candidate statements against real decisions including one they should reject. | ### Build and maintain | Skill | Use it for | | --- | --- | | [ADX](packages/adx/README.md) | Give coding agents checked repository commands, verification and recovery, with drift checks and a small-repository path. | | [App It](packages/app-it/README.md) | Make a local web project a macOS app with a native window, managed servers, warm reopen and explicit cleanup. | | [Autoloop](packages/autoloop/README.md) | Plan and run finite local file experiments that measure the original first and retain only guarded improvements in a separate Git repository. | | [Ghost Invasion](packages/ghost-invasion/README.md) | Run seeded browser journeys against a disposable local app and reproduce failures with captured evidence. | | [Kit the Skill](packages/kit-the-skill/README.md) | Decides whether a skill needs supporting files at all, and if so the smallest kind that changes what the agent does, on a ladder from no addition up to executable support. | | [Secrets Rotation](packages/secrets-rotation/README.md) | Rehearse staged credential replacement, consumer reload checks, grace, revocation and recovery with a local fake provider. | ### Presentation and walkthroughs | Skill | Use it for | | --- | --- | | [Product Walkthrough](packages/product-walkthrough/README.md) | Record expectations before acting, expose broken promises in a running product, and validate findings, evidence and honest coverage. | | [Repo Craft](packages/repo-craft/README.md) | Turn repository presentation into useful proof of the actual project, with a portable probe and reproducible terminal visuals. | ### Feedback and improvement | Skill | Use it for | | --- | --- | | [Agent Coach](packages/agent-coach/README.md) | Analyze selected sessions, track applied advice separately, and observe whether its associated problem changes in later comparable sessions. | | [Closed Loop](packages/closed-loop/README.md) | Finds where a measured outcome should change a later action and wires the smallest feedback loop that does it, or says to stay stateless. | | [Health Synthesis](packages/health-synthesis/README.md) | Combine review findings while preserving each ID, recorded verdict and conflict outcome; check coverage and merge consistency. | | [Token Optimizer](packages/token-optimizer/README.md) | Find and repair repeated context, retrieval, caching, tool-definition and retry waste, measuring whole-task behavior before claiming savings. | ## Inspect the evidence - [Repo Craft's Fairshare CLI](packages/repo-craft/examples/walkthrough.md): executed terminal output, repository before and after, and editable image sources. The project is synthetic; the commands ran. - [Product Walkthrough's Pocket List](packages/product-walkthrough/examples/walkthrough.md): a note lost on reload, a passing delete-cancellation check, screenshots and a runnable app. Coverage is deliberately partial. - [Ghost Invasion's double-submit case](packages/ghost-invasion/examples/walkthrough.md): seeded failure, repeat, reduced reproduction and fixed control with browser and database evidence. It needs its bundled engine, a browser and Docker. - [Secrets Rotation's offline tutorial](packages/secrets-rotation/examples/walkthrough.md): setup, two rotations, failed rollout and recovery using only generated fake credentials and a fake provider. - [Vision's repaired comparison](packages/vision/comparison/difficult/evaluation.md): the original difficult result and rerun remain inspectable. Only an authorized decision-maker's approval establishes an intentional pivot. - [Thinking Hats' migration comparison](packages/thinking-hats/comparison/difficult/evaluation.md): narrow additional retirement and rollback decisions. Both answers already catch the main integrity problems; no migration was executed. Each package retains its full comparison record, including unfavorable results and exclusions. The [collection's own Repo Craft report](docs/collection-presentation/report.md) records how this landing page was assembled. ## Verify or build locally ```sh . .venv/bin/activate python3 -B tools/validate_packages.py --packages-root packages --catalog site/catalog.json --release python3 -B tools/verify_release.py cd site npm ci npm run build:release ``` The release check installs and reverses every package in disposable Codex and Claude directories. Package-specific commands and conditions live beside each package. The static site requires Node 22.12 or newer and supports root and project-subpath builds. See [site setup and preview](site/README.md) and the [complete release verification commands](docs/verification.md). ## Reuse and contribute Original collection adaptations use [MIT](LICENSE). [Credits and notices](NOTICE.md) preserve upstream material and acknowledge established methods; the collection does not claim their invention. Dependencies keep their own licenses. Read [support and limitations](SUPPORT.md) and [contribution guidance](CONTRIBUTING.md). No installation command launches a model, configures a paid provider, or deploys the site.

Carregando o README do repositório…

Skills neste repositório

Outras ocupações de informática Desenvolvedores de software Especialistas em gestão de projetos Administradores de redes e sistemas de computador

8 categorias ocupacionais · 100% classificado

skill

ocupação

descrição

atualizado

adx-forensic

Administradores de redes e sistemas de computador

Diagnose wrong setup commands, unclear verification, recovery gaps and stale repository guidance. Use for first-time agent onboarding or an evidence-backed ADX assessment. Includes the executable ADX kit; no companion installation is required.