# AI Library: Agent Skills and Plugins

> Skills I use with AI coding agents, ready to read, copy and download. Originals are published here in full; curated ones link to their authors, with what I changed and why I run them.

Canonical: https://mikereams.com/ai

## Start here

- **Map a portfolio**: From a list of applications to a capability map in your own tool. [catalog-lookup](https://mikereams.com/ai/catalog-lookup) → [capability-mapping](https://mikereams.com/ai/capability-mapping) → [product-coverage](https://mikereams.com/ai/product-coverage) → [export-to-tool](https://mikereams.com/ai/export-to-tool)
- **Stress-test a plan**: Before a plan is approved. [grill-me](https://mikereams.com/ai/grill-me) → [pre-mortem](https://mikereams.com/ai/pre-mortem) → [strategy-red-team](https://mikereams.com/ai/strategy-red-team)
- **Ship a change with an agent**: From the first test to the merge. [test-driven-development](https://mikereams.com/ai/test-driven-development) → [verification-before-completion](https://mikereams.com/ai/verification-before-completion) → [code-review](https://mikereams.com/ai/code-review) → [receiving-code-review](https://mikereams.com/ai/receiving-code-review) → [security-reviewer](https://mikereams.com/ai/security-reviewer)
- **Before you install a skill**: For a skill from a source you do not control. [skill-intake](https://mikereams.com/ai/skill-intake) → [skill-security-auditor](https://mikereams.com/ai/skill-security-auditor)

## Original

### Blueprint Catalog plugin

Plugin · Original · Enterprise architecture · updated 2026-11-13. Details: https://mikereams.com/ai/blueprint-catalog

The nine Blueprint Catalog skills and the catalog's read-only MCP server, installed together as one Claude Code plugin.

What you get: Nine enterprise architecture skills ready to use, with the catalog connected and nothing to configure.

When to use it: When you use Claude Code and want the catalog skills without setting up the MCP server yourself.

Ask for it like this: "Look up incident management in the catalog and give me its id and scope."

Limits: Claude Code only. Other agents install the skills one at a time with the skills CLI and connect the server themselves.

- It adds about 650 tokens to every session; a skill costs more only when it runs.
- The server is read-only and needs no key.

Installs 9 skills ([catalog-lookup](https://mikereams.com/ai/catalog-lookup), [capability-mapping](https://mikereams.com/ai/capability-mapping), [product-coverage](https://mikereams.com/ai/product-coverage), [value-stream-worksheet](https://mikereams.com/ai/value-stream-worksheet), [export-to-tool](https://mikereams.com/ai/export-to-tool), [adr-with-patterns](https://mikereams.com/ai/adr-with-patterns), [lifecycle-check](https://mikereams.com/ai/lifecycle-check), [industry-brief](https://mikereams.com/ai/industry-brief), [calm-tagging](https://mikereams.com/ai/calm-tagging)) and one read-only MCP server, https://catalog.mikereams.com/mcp (CC BY 4.0).

### Capability mapping and heat map

Skill · Original · Enterprise architecture · updated 2026-10-29. Details: https://mikereams.com/ai/capability-mapping

Maps a list of applications, products or team responsibilities onto business capabilities, flags the ambiguous ones, and fills a heat-map CSV.

What you get: A capability heat map (template TM-20) with every item placed at the right level, ids kept for re-import, and a list of what could not be placed.

When to use it: For application portfolio assessments, CMDB business-capability modeling and heat maps.

Ask for it like this: "Map these 40 applications to business capabilities and give me the heat map."

Limits: It will not name a capability after an application or mirror the org chart; what it cannot place, it lists.

- Written for the Blueprint Catalog, my free capability catalog; every path, tool and id it names is checked against each catalog build.
- Searches by what an item does, not its product name, and warns against capabilities named after applications or the org chart.

On the Blueprint Catalog (CC BY 4.0): [SKILL.md](https://catalog.mikereams.com/skills/capability-mapping/SKILL.md). Needs: The catalog's read-only MCP server (catalog.mikereams.com/mcp) or its HTTPS API. No key, no install.

### Product coverage comparison

Skill · Original · Enterprise architecture · updated 2026-10-29. Details: https://mikereams.com/ai/product-coverage

Compares how two or more products cover a set of business capabilities, with overlaps, gaps and the catalog's maturity and confidence.

What you get: A capability-by-product table, then the gaps nobody covers and the overlaps where two products both claim strength.

When to use it: For build-versus-buy, platform consolidation and vendor shortlists.

Ask for it like this: "Compare the products that cover service desk and asset management, and show me the gaps."

Limits: Low confidence stays a lead to check with the vendor, and it claims no affiliation with any vendor.

- Written for the Blueprint Catalog, my free capability catalog; every path, tool and id it names is checked against each catalog build.
- Low confidence is treated as a lead to check with the vendor, not a fact.

On the Blueprint Catalog (CC BY 4.0): [SKILL.md](https://catalog.mikereams.com/skills/product-coverage/SKILL.md). Needs: The catalog's read-only MCP server (catalog.mikereams.com/mcp) or its HTTPS API. No key, no install.

### Value stream worksheet

Skill · Original · Enterprise architecture · updated 2026-10-29. Details: https://mikereams.com/ai/value-stream-worksheet

Drafts an organization's value stream from the nearest catalog stream, then fills the worksheet with its own stages, pain points and capabilities.

What you get: A filled value stream worksheet (template TM-30): four to eight stages, owners, pain points and the capabilities each stage leans on.

When to use it: When an organization needs its own value stream, drafted from the nearest one in the catalog.

Ask for it like this: "Draft our order-to-cash value stream and fill in the worksheet."

- Written for the Blueprint Catalog, my free capability catalog; every path, tool and id it names is checked against each catalog build.
- Uses each stage's domain events as the test of a stage boundary.

On the Blueprint Catalog (CC BY 4.0): [SKILL.md](https://catalog.mikereams.com/skills/value-stream-worksheet/SKILL.md). Needs: The catalog's read-only MCP server (catalog.mikereams.com/mcp) or its HTTPS API. No key, no install.

### Export a capability map to your tool

Skill · Original · Enterprise architecture · updated 2026-10-29. Details: https://mikereams.com/ai/export-to-tool

Gets the capability map into a CMDB, an EA tool or an ArchiMate model, with the right export file and its import notes.

What you get: The export file shaped for the target tool, the import steps, and the catalog version recorded for the next refresh.

When to use it: When loading or refreshing a capability model in an EA or CMDB tool.

Ask for it like this: "Get the capability map into our CMDB and give me the import steps."

Limits: It prepares the file and the steps; the import itself is yours, on a non-production instance first.

- Written for the Blueprint Catalog, my free capability catalog; every path, tool and id it names is checked against each catalog build.
- Rows come parents first, and a refresh reads the change log so merged records are retired instead of duplicated.

On the Blueprint Catalog (CC BY 4.0): [SKILL.md](https://catalog.mikereams.com/skills/export-to-tool/SKILL.md). Needs: The catalog's read-only MCP server (catalog.mikereams.com/mcp) or its HTTPS API. No key, no install.

### Decision records that cite patterns

Skill · Original · Enterprise architecture · updated 2026-10-29. Details: https://mikereams.com/ai/adr-with-patterns

Writes an architecture decision record that names the design patterns behind each option, the anti-patterns it avoids and the capability it serves.

What you get: An ADR (template TM-10) with every pattern and anti-pattern cited by id and link.

When to use it: When recording a design choice, a trade-off or a standard.

Ask for it like this: "Write the decision record for choosing event streaming over batch, with the patterns behind each option."

Limits: A pattern whose trade-off the record cannot state is left out.

- Written for the Blueprint Catalog, my free capability catalog; every path, tool and id it names is checked against each catalog build.
- If the record cannot state the trade-off a pattern answers, the pattern is decoration.

On the Blueprint Catalog (CC BY 4.0): [SKILL.md](https://catalog.mikereams.com/skills/adr-with-patterns/SKILL.md). Needs: The catalog's read-only MCP server (catalog.mikereams.com/mcp) or its HTTPS API. No key, no install.

### Technology lifecycle check

Skill · Original · Enterprise architecture · updated 2026-10-29. Details: https://mikereams.com/ai/lifecycle-check

Reports which release cycles in a stack are still maintained, which have ended and what ends next, from the catalog's endoflife.date snapshot.

What you get: A table per technology: the version in use, its end of life, extended support, and the current and LTS cycles.

When to use it: For technical debt reviews, upgrade planning and risk registers.

Ask for it like this: "Check the lifecycle of our stack: PostgreSQL 14, .NET 8 and Node 20."

Limits: It judges against the date of the catalog's snapshot, not today, and says when that was.

- Written for the Blueprint Catalog, my free capability catalog; every path, tool and id it names is checked against each catalog build.
- Judges against the snapshot date, not today, and says when the snapshot was taken.

On the Blueprint Catalog (CC BY 4.0): [SKILL.md](https://catalog.mikereams.com/skills/lifecycle-check/SKILL.md). Needs: The catalog's read-only MCP server (catalog.mikereams.com/mcp) or its HTTPS API. No key, no install.

### Industry brief

Skill · Original · Enterprise architecture · updated 2026-10-29. Details: https://mikereams.com/ai/industry-brief

Produces a one-page brief on how the catalog covers an industry: capabilities, value streams, products, technology and governing standards.

What you get: One page on a sector, with headline counts and the review status stated plainly.

When to use it: To orient on a sector before a mapping or an assessment.

Ask for it like this: "Give me the one-page brief for healthcare."

Limits: It reports how the catalog covers the sector, with the review status stated; it is not a study of the industry.

- Written for the Blueprint Catalog, my free capability catalog; every path, tool and id it names is checked against each catalog build.
- Useful before a mapping or assessment in a sector you are new to.

On the Blueprint Catalog (CC BY 4.0): [SKILL.md](https://catalog.mikereams.com/skills/industry-brief/SKILL.md). Needs: The catalog's read-only MCP server (catalog.mikereams.com/mcp) or its HTTPS API. No key, no install.

### Tag a FINOS CALM architecture

Skill · Original · Enterprise architecture · updated 2026-10-29. Details: https://mikereams.com/ai/calm-tagging

Tags a CALM architecture's nodes, relationships and flows with the business capabilities, value stream stages and data concepts they support.

What you get: A CALM 1.2 file whose elements carry checked capability, stage and data-concept ids, plus a list of what was left untagged and why.

When to use it: When a CALM model should be read against a shared business model.

Ask for it like this: "Tag this CALM architecture file with the capabilities and value stream stages it supports."

Limits: Every id is checked against the catalog; what it leaves untagged, it lists with the reason.

- Written for the Blueprint Catalog, my free capability catalog; every path, tool and id it names is checked against each catalog build.
- Two nodes that both master the same data concept is a finding worth raising.

On the Blueprint Catalog (CC BY 4.0): [SKILL.md](https://catalog.mikereams.com/skills/calm-tagging/SKILL.md). Needs: The catalog's read-only MCP server (catalog.mikereams.com/mcp) or its HTTPS API. No key, no install.

### Catalog lookup

Skill · Original · Enterprise architecture · updated 2026-10-29. Details: https://mikereams.com/ai/catalog-lookup

Finds a capability, product, pattern, technology or glossary term by name or alias and returns its id, definition, scope and canonical link.

What you get: The catalog's record for a term: id, definition, what is in and out of scope, and the page to link.

When to use it: Before citing or linking any catalog entry.

Ask for it like this: "Look up incident management in the catalog and give me its id and scope."

Limits: It quotes the catalog's definition only with attribution; otherwise it paraphrases.

- Written for the Blueprint Catalog, my free capability catalog; every path, tool and id it names is checked against each catalog build.
- The other catalog skills start here, so terms and ids stay consistent.

On the Blueprint Catalog (CC BY 4.0): [SKILL.md](https://catalog.mikereams.com/skills/catalog-lookup/SKILL.md). Needs: The catalog's read-only MCP server (catalog.mikereams.com/mcp) or its HTTPS API. No key, no install.

### Technology request intake

Skill · Original · Enterprise architecture · updated 2026-11-12. Details: https://mikereams.com/ai/technology-request-intake

Asks every "can we buy X?" request the same six questions, exposes overlap with tools already owned, and decides which records the request creates.

What you get: A review summary written for the request itself: six answers, a net-new or overlap verdict, a recommendation and the records to create.

When to use it: For software, SaaS and tool requests, before anyone signs anything.

Ask for it like this: "Review this request for a new survey tool against our approved products and application list."

Limits: It recommends and a person decides. What it could not check against your portfolio is marked not checked, never guessed.

- Written from two posts in my application portfolio series: the six questions, and the rule for which record a request becomes.
- Question three does the work: one primary technical class, the incumbents already in that branch, and a verdict.

Files (MIT): [SKILL.md](https://mikereams.com/ai-files/technology-request-intake/SKILL.md)

### Code review with a coverage ledger

Skill · Original · Review & QA · updated 2026-10-17. Details: https://mikereams.com/ai/code-review

Reviews a change file by file, records every file as reviewed or skipped with a reason, and reports only findings it has re-checked.

What you get: A ledger of every changed file and a short, severity-ranked list of confirmed findings with line numbers.

When to use it: Before a commit, a merge or a milestone sign-off: a working copy, a commit, a branch range or a PR.

Ask for it like this: "Review this branch against main before I merge it."

Limits: Read-only by default, and your project's own instructions win over its rules.

- Built to stop two failures: a reviewer who reads 6 of 14 files and says it looks good, and twenty nitpicks that bury the one real defect.
- A check mode compares the app's own docs with its code and lists only the disagreements, for a release gate.
- The estate rules treat a new dependency, a secret in code or work content in a personal repo as stop-and-ask.

Files (MIT for the original parts; bundled third-party parts keep their licences (see NOTICE); Reviewer prompt adapted from obra/superpowers (MIT); rule docs from alibaba/open-code-review (Apache-2.0).): [SKILL.md](https://mikereams.com/ai-files/code-review/SKILL.md), [reviewer-prompt.md](https://mikereams.com/ai-files/code-review/reviewer-prompt.md), [rules/default.md](https://mikereams.com/ai-files/code-review/rules/default.md), [rules/estate.md](https://mikereams.com/ai-files/code-review/rules/estate.md), [rules/ts-js.md](https://mikereams.com/ai-files/code-review/rules/ts-js.md), [rules/github-workflows.md](https://mikereams.com/ai-files/code-review/rules/github-workflows.md), [rules/package-json.md](https://mikereams.com/ai-files/code-review/rules/package-json.md), [NOTICE](https://mikereams.com/ai-files/code-review/NOTICE), [LICENSE-MIT-superpowers](https://mikereams.com/ai-files/code-review/LICENSE-MIT-superpowers), [LICENSE-APACHE-2.0-open-code-review](https://mikereams.com/ai-files/code-review/LICENSE-APACHE-2.0-open-code-review)

### Skill intake

Skill · Original · Security · updated 2026-11-14. Details: https://mikereams.com/ai/skill-intake

Asks four questions before a skill is installed (who verified it, what can it run, which version, where does it run) and records the verdict.

What you get: A verdict (accept, accept after edits or reject) with the evidence for it, and one row in a review record you can audit later.

When to use it: Before installing, vendoring or updating a skill, a plugin or an MCP server from a source you do not control.

Ask for it like this: "Run skill intake on this repository's skills folder before I install anything."

Limits: It reads; it never installs or runs the skill, and it does not follow instructions found inside it.

- Written from my post on vendor skill catalogs: the four questions there, turned into a procedure.
- The review record is the point. Every curated skill in this library has a row like the one it writes.

Files (MIT): [SKILL.md](https://mikereams.com/ai-files/skill-intake/SKILL.md)

### Diagrams: Mermaid or blueprint SVG

Skill · Original · Diagrams · updated 2026-10-28. Details: https://mikereams.com/ai/diagrams

Picks Mermaid or a hand-drawn blueprint SVG based on where the diagram will live, then draws flowcharts, sequences, ERDs, C4 views and state machines.

What you get: A diagram that renders where it is going: a Mermaid block for GitHub or Obsidian, or a clean grid-aligned SVG for docs and slides.

When to use it: When asked to diagram, visualize or map out a flow, a user journey, an architecture or a state machine.

Ask for it like this: "Draw the sign-up flow as a sequence diagram I can keep in the repo."

Limits: One concept per diagram: it splits a crowded one instead of cramming it.

- One concept per diagram: split rather than cram.
- Component state charts pair with the States section of a design handoff spec.

Files (MIT): [SKILL.md](https://mikereams.com/ai-files/diagrams/SKILL.md)

### ServiceNow write check

Skill · Original · Enterprise architecture · updated 2026-11-14. Details: https://mikereams.com/ai/servicenow-write-check

Stops an agent trusting a success code from the ServiceNow Table API: check the field exists, write the stored value, read every write back.

What you get: A sent-versus-stored comparison for every write, and a pass or fail on whether each field exists at all.

When to use it: Before a load, an update script or a report against a ServiceNow table, and when a write returned success but nothing changed.

Ask for it like this: "Check that this field exists on the table, then update one record and read it back before the batch."

Limits: It never writes to production unless you ask for that write, and it stops at the first mismatch.

- Written from two posts: five Table API calls that return success and store nothing, and the partition test for a field that is not there.
- Instances differ, so each rule is a test to run on yours. It uses the public Table and Aggregate APIs only.

Files (MIT): [SKILL.md](https://mikereams.com/ai-files/servicenow-write-check/SKILL.md)

### UI critique

Skill · Original · Frontend · updated 2026-11-13. Details: https://mikereams.com/ai/ui-critique

Decides what a page must do before it is designed, then judges what was built from screenshots at three widths, with measured checks and a score.

What you get: Screenshots at three widths, a measured report, a scored critique with ranked findings, and the top three fixed, for up to three rounds.

When to use it: Before designing a public page, after building any page or app screen, and before a UI milestone sign-off.

Ask for it like this: "Critique the pricing page: capture it, score it and fix the top three."

Limits: Local URLs by default and read-only towards the page. Passing the measured checks says nothing about whether the page is good.

- It judges the pixels, not the code: screenshots at 1440, 768 and 390 pixels, then a fixed set of passes.
- The measured checks catch mechanical problems only. The scored critique is where the judgement is.
- It uses the Playwright the app already has and adds no dependency.

Files (MIT; Method adapted from Leonxlnx/taste-skill (MIT), with ideas from pbakaus/impeccable (Apache-2.0). No text or code was copied.): [SKILL.md](https://mikereams.com/ai-files/ui-critique/SKILL.md), [references/rubric.md](https://mikereams.com/ai-files/ui-critique/references/rubric.md), [references/tells.md](https://mikereams.com/ai-files/ui-critique/references/tells.md), [references/argument.md](https://mikereams.com/ai-files/ui-critique/references/argument.md), [references/app-screens.md](https://mikereams.com/ai-files/ui-critique/references/app-screens.md), [references/moves.md](https://mikereams.com/ai-files/ui-critique/references/moves.md), [references/harden.md](https://mikereams.com/ai-files/ui-critique/references/harden.md), [scripts/critique.mjs](https://mikereams.com/ai-files/ui-critique/scripts/critique.mjs), [scripts/probe.js](https://mikereams.com/ai-files/ui-critique/scripts/probe.js), [scripts/record.mjs](https://mikereams.com/ai-files/ui-critique/scripts/record.mjs), [scripts/thresholds.json](https://mikereams.com/ai-files/ui-critique/scripts/thresholds.json), [NOTICE](https://mikereams.com/ai-files/ui-critique/NOTICE)

## Curated

### Web interface guidelines

Skill · Curated · Frontend · updated 2026-08-20. Details: https://mikereams.com/ai/web-design-guidelines

Checks a UI against a long list of interface rules for accessibility, forms, focus, motion and copy.

What you get: A list of rule violations in the UI, each tied to the element that breaks it.

When to use it: When asked to review a UI, check accessibility or audit a design.

Ask for it like this: "Review this page's UI against the interface guidelines."

Limits: It never fetches rules at run time; the rule set is the copy in the folder.

- The original downloaded its rules from GitHub on every run and followed the output instructions it found there: a live remote-instruction channel.
- My copy keeps the rules as a local file and never fetches anything.
- A second reader for forms, focus and motion rules that an automated accessibility scan does not cover.

Source: [vercel-labs/agent-skills](https://github.com/vercel-labs/agent-skills/tree/b8caa26/skills/web-design-guidelines) by Vercel Labs (MIT), reviewed at b8caa26. Not hosted here.

### Web best practices (Lighthouse)

Skill · Curated · Frontend · updated 2026-08-20. Details: https://mikereams.com/ai/frontend-best-practices

Runs a Lighthouse-style best-practices audit: security headers, CSP, deprecated APIs, console errors and valid markup.

What you get: A checked-off audit list for a build, with the failures named.

When to use it: As the last gate before an app is called done.

Ask for it like this: "Run the best-practices audit on this build."

- Condensed from Addy Osmani's best-practices skill; the checks are his.
- I run a condensed checklist version as the last gate before an app is called done.

Source: [addyosmani/web-quality-skills](https://github.com/addyosmani/web-quality-skills/tree/main/skills/best-practices) by Addy Osmani (MIT). Not hosted here.

### UI design system toolkit

Skill · Curated · Design systems · updated 2026-08-20. Details: https://mikereams.com/ai/ui-design-system

Generates design tokens from a brand colour, checks contrast, does responsive type maths and writes developer handoff specs.

What you get: A token set (colour ramps, type scale, spacing, motion) and a handoff spec a developer can build from without asking a question.

When to use it: When creating a design system or keeping one consistent.

Ask for it like this: "Generate a token set from this brand colour and check the contrast of every text pair."

Limits: It runs the generator; it does not hand-write a token set.

- This site's own tokens came out of its generator, then a contrast pass on every text pair.
- I run a condensed version of the skill.

Source: [alirezarezvani/claude-skills](https://github.com/alirezarezvani/claude-skills) by Alireza Rezvani (MIT). Not hosted here.

### Editorial diagram design

Skill · Curated · Diagrams · updated 2026-09-26. Details: https://mikereams.com/ai/diagram-design

Draws diagrams with an editorial eye: templates, a style guide and layout rules that make technical drawings readable.

What you get: A polished, on-brand technical diagram in HTML or SVG.

When to use it: For a diagram that will be presented: architecture, current state, sequence, swimlane and more.

Ask for it like this: "Make a presentation-quality current-state diagram of this system."

Limits: Offline by default: no web fonts, and no profile files outside the project.

- Offline fonts by default: I removed the Google Fonts links from the templates.
- Replaced an unpinned third-party browser CLI, and removed icons whose licences I could not verify.
- For diagrams that go in front of people, where layout and type matter more than speed. My own diagrams skill covers the quick ones.

Source: [cathrynlavery/diagram-design](https://github.com/cathrynlavery/diagram-design/tree/dc1ace4/skills/diagram-design) by Cathryn Lavery (MIT), reviewed at dc1ace4. Not hosted here.

### Archify

Skill · Curated · Diagrams · updated 2026-09-26. Details: https://mikereams.com/ai/archify

Builds architecture, workflow, sequence, data-flow and state diagrams as one standalone HTML page, from a small typed specification it validates first.

What you get: An explorable diagram in a single HTML file, with dark and light themes and export to PNG, SVG and more.

When to use it: When an architecture or a flow should be explored on screen or exported, not only read in a document.

Ask for it like this: "Draw the data flow for this service as a standalone HTML diagram."

Limits: My copy never calls the network: no update check, and no brand marks fetched at render time.

- I removed its update check, which called the author's site every 72 hours, and the scripts that went with it.
- For a diagram someone will click through: it validates the layout before it hands the file over. My own diagrams skill covers the quick ones.
- MIT at the commit I reviewed, and built on Cocoon AI's architecture diagram generator. The project has since moved to Apache-2.0, so check the licence if you take a newer version.

Source: [tt-a1i/archify](https://github.com/tt-a1i/archify/tree/9e35d2b/archify) by tt-a1i (MIT), reviewed at 9e35d2b. Not hosted here.

### Systematic debugging

Skill · Curated · Debugging · updated 2026-10-01. Details: https://mikereams.com/ai/systematic-debugging

Finds the root cause before touching code: reproduce, trace back to the source, form one hypothesis at a time, then fix.

What you get: A root cause with the evidence for it, and a fix at the source instead of a patch where the symptom showed.

When to use it: On any bug, test failure or unexpected behaviour, before proposing a fix.

Ask for it like this: "This test fails now and then; find the root cause before changing anything."

Limits: No fix until the failure is reproduced; if it cannot be, it gathers more data instead of guessing.

- I dropped the evaluation fixtures and a shell script that ran an unscoped test command.
- My rule is to reproduce a bug and find its root cause before patching. This is the procedure for that rule.

Source: [obra/superpowers](https://github.com/obra/superpowers/tree/8ca22db/skills/systematic-debugging) by Jesse Vincent (MIT), reviewed at 8ca22db. Not hosted here.

### Verification before completion

Skill · Curated · Review & QA · updated 2026-10-01. Details: https://mikereams.com/ai/verification-before-completion

No 'done', 'fixed' or 'passing' without fresh evidence: run the check, read the output, then make the claim.

What you get: Every completion claim backed by a command run in the same session and its output.

When to use it: Before claiming work is complete, fixed or passing, and before a commit or a PR.

Ask for it like this: "Before you call it done, show me the command and its output."

Limits: No success claim without a command run in the same session.

- Unchanged. Text only, no scripts.
- My agents may not call anything done without evidence. This skill turns that rule into a habit.

Source: [obra/superpowers](https://github.com/obra/superpowers/tree/8ca22db/skills/verification-before-completion) by Jesse Vincent (MIT), reviewed at 8ca22db. Not hosted here.

### Test-driven development

Skill · Curated · Review & QA · updated 2026-10-01. Details: https://mikereams.com/ai/test-driven-development

Write the failing test, watch it fail for the right reason, write the least code that passes, then refactor.

What you get: New behaviour that arrives with a test that failed without it.

When to use it: When implementing any feature or bug fix, before writing the implementation.

Ask for it like this: "Add this validation rule test-first."

Limits: Code written before its test is deleted, not kept as a reference.

- Wording edits only: names and one cross-reference.
- For anything testable, I want the failing test written and seen to fail first.

Source: [obra/superpowers](https://github.com/obra/superpowers/tree/8ca22db/skills/test-driven-development) by Jesse Vincent (MIT), reviewed at 8ca22db. Not hosted here.

### Intended vs implemented

Skill · Curated · Review & QA · updated 2026-10-17. Details: https://mikereams.com/ai/intended-vs-implemented

Compares what an app's docs say it does with what the code actually enforces, and cites both sides for every gap.

What you get: A list of gaps between documented intent and the code, each with the doc line, the code line and who could exploit it.

When to use it: When documented intent exists: a spec, decision records or project rules.

Ask for it like this: "Compare the spec's permission rules with what the code enforces."

Limits: It reports only the mismatches that matter: ones that let someone reach data, money or infrastructure they should not.

- Text only. I pointed it at my own spec and decision files instead of the author's document set.
- I run it at the end of each milestone, so that 'accepted' means the spec and the code agree.
- The author has since moved the skill to phuryn/product-engineering. The link here is still the version I reviewed.

Source: [phuryn/pm-skills](https://github.com/phuryn/pm-skills/tree/07b34f6/pm-ai-shipping/skills/intended-vs-implemented) by Paweł Huryn (MIT), reviewed at 07b34f6. Not hosted here.

### Improve codebase architecture

Skill · Curated · Review & QA · updated 2026-10-29. Details: https://mikereams.com/ai/improve-codebase-architecture

Scans a codebase for shallow modules, shows the candidates in a visual report, then interviews you about the one you pick.

What you get: A report of the places where a refactor would put more behaviour behind a smaller interface, and a worked plan for the one you choose.

When to use it: When a codebase is hard to test or hard to find your way around, and before a larger refactor.

Ask for it like this: "Find the deepening opportunities in this codebase and show me the report."

Limits: My copy writes the report to the temp folder, not the repository, and gives you the path instead of opening it.

- Two skills, one card: codebase-design is the vocabulary the review speaks in (module, interface, depth, seam).
- I made the report self-contained. Upstream it loaded two scripts from CDNs, with a loose diagram setting, and opened the file with a shell command; mine loads nothing.
- I run it when the same files keep coming up in fixes. It starts from the commit history, so it looks where the change is.

Source: [mattpocock/skills](https://github.com/mattpocock/skills/tree/59db609/skills/engineering/improve-codebase-architecture) by Matt Pocock (MIT), reviewed at 59db609. Not hosted here. Also covers [codebase-design](https://github.com/mattpocock/skills/tree/59db609/skills/engineering/codebase-design), at the same commit.

### Receiving code review

Skill · Curated · Review & QA · updated 2026-10-01. Details: https://mikereams.com/ai/receiving-code-review

Treats review feedback as a technical claim to check: restate it, verify it against the code, then implement it or push back.

What you get: Each review comment either implemented and tested, one at a time, or answered with the reason it does not apply here.

When to use it: When review feedback arrives, before implementing any of it, above all when a comment is unclear or looks wrong.

Ask for it like this: "Here is the review on my PR; check each comment against the code before changing anything."

Limits: No performative agreement, and nothing is implemented while any comment is still unclear.

- Wording edits only: names, and a note on how replies are posted to GitHub.
- The other half of code review. An agent that agrees with every comment is as little use as a reviewer who approves every change.

Source: [obra/superpowers](https://github.com/obra/superpowers/tree/8ca22db/skills/receiving-code-review) by Jesse Vincent (MIT), reviewed at 8ca22db. Not hosted here.

### Grill me

Skill · Curated · Planning · updated 2026-10-29. Details: https://mikereams.com/ai/grill-me

Interviews you about a plan in rounds, with a recommended answer for every question, until no decision is left assumed.

What you get: A plan whose decisions have all been settled out loud, in the order they depend on each other.

When to use it: When a plan, a design or a decision needs stress-testing before work starts.

Ask for it like this: "Grill me on this migration plan."

Limits: It looks facts up itself instead of asking you for them, and it does not act on the plan until you confirm.

- Two skills, one card: grill-me is the command and grilling holds the method. Text only; I changed the round format to plain labels instead of emoji.
- It runs on a milestone plan before the plan reaches me for approval, next to the pre-mortem. The questions find what the plan assumed.

Source: [mattpocock/skills](https://github.com/mattpocock/skills/tree/59db609/skills/productivity/grill-me) by Matt Pocock (MIT), reviewed at 59db609. Not hosted here. Also covers [grilling](https://github.com/mattpocock/skills/tree/59db609/skills/productivity/grilling), at the same commit.

### Grill with docs

Skill · Curated · Planning · updated 2026-10-29. Details: https://mikereams.com/ai/grill-with-docs

The same interview, with the answers written down as it goes: terms into a glossary, decisions into decision records.

What you get: A settled plan, plus a glossary and decision records written while the answers are fresh.

When to use it: When a plan introduces terms or decisions that the next person will need to find.

Ask for it like this: "Grill me on this design and write the glossary and the decision records as we go."

Limits: It creates a file only when there is something to write: no empty glossary, no placeholder records.

- Two skills, one card: grill-with-docs runs the interview and domain-modeling, which keeps the glossary and the records. Text only; I added where terms and records go in my own notes.
- The closest of the set to architecture work: the interview ends with records, not only with agreement.

Source: [mattpocock/skills](https://github.com/mattpocock/skills/tree/59db609/skills/engineering/grill-with-docs) by Matt Pocock (MIT), reviewed at 59db609. Not hosted here. Also covers [domain-modeling](https://github.com/mattpocock/skills/tree/59db609/skills/engineering/domain-modeling), at the same commit.

### Pre-mortem

Skill · Curated · Planning · updated 2026-10-17. Details: https://mikereams.com/ai/pre-mortem

Imagines the plan has already failed and works back: real threats, paper tigers and the elephants nobody mentions.

What you get: A ranked list of what is most likely to sink the plan, with the mitigations worth doing now.

When to use it: On a milestone plan, a spec or a launch plan, before approving it.

Ask for it like this: "Run a pre-mortem on this milestone plan."

- Text only. "Launch" now also means a milestone, a release or a handoff, and it reads the plan I name.
- Cheap to run on a milestone plan before I approve it.

Source: [phuryn/pm-skills](https://github.com/phuryn/pm-skills/tree/07b34f6/pm-execution/skills/pre-mortem) by Paweł Huryn (MIT), reviewed at 07b34f6. Not hosted here.

### Strategy red team

Skill · Curated · Planning · updated 2026-10-17. Details: https://mikereams.com/ai/strategy-red-team

Attacks a strategy or plan the way a sceptic would, and ranks the objections by the cheapest test that would settle each one.

What you get: The weakest assumptions in a plan, each with the cheapest experiment that would prove or kill it.

When to use it: On a plan, a roadmap or a strategy, before reality tests it.

Ask for it like this: "Red-team this roadmap and give me the cheapest test for each weak assumption."

Limits: It does not invent risks: a claim that is well reasoned is said to be.

- I removed its cross-model step, which sent the plan to other AI services.
- Most useful where the risk is building the wrong thing well.

Source: [phuryn/pm-skills](https://github.com/phuryn/pm-skills/tree/07b34f6/pm-execution/skills/strategy-red-team) by Paweł Huryn (MIT), reviewed at 07b34f6. Not hosted here.

### Handoff

Skill · Curated · Planning · updated 2026-10-29. Details: https://mikereams.com/ai/handoff

Compacts the current session into a handoff document, so a fresh agent can continue the work without the conversation.

What you get: One document that says where the work stands and which skills the next session should load, pointing at specs and commits instead of repeating them.

When to use it: At the end of a session, or when a conversation has grown too long to continue in.

Ask for it like this: "Write a handoff for the next session; it will finish the import step."

Limits: It redacts keys, passwords and personal details, and it refers to existing specs and commits instead of copying them.

- Text only. I changed where the document is saved, so it never lands in an app's repository, and added a rule that client work stays in its own workspace.
- Sessions forget. A handoff is the cheapest way I know to make the next one start where this one stopped.

Source: [mattpocock/skills](https://github.com/mattpocock/skills/tree/59db609/skills/productivity/handoff) by Matt Pocock (MIT), reviewed at 59db609. Not hosted here.

### Security reviewer

Skill · Curated · Security · updated 2026-10-17. Details: https://mikereams.com/ai/security-reviewer

Reviews code for security defects: injection, authentication and authorisation gaps, secrets and unsafe defaults.

What you get: Security findings ranked by severity, each with the affected code and a fix.

When to use it: For a security audit, or a review of code for vulnerabilities.

Ask for it like this: "Review the sign-in and upload code for security defects."

Limits: My copy reads and searches code; it does not run commands or test live systems.

- I removed its penetration-testing reference, which included commands that read home directories and shell history.
- My copy can read and search code but not run commands.
- A second, security-only pass after code review, on anything that handles input, sign-in or secrets.

Source: [Jeffallan/claude-skills](https://github.com/Jeffallan/claude-skills/tree/882ef55/skills/security-reviewer) by Jeff Allan (MIT), reviewed at 882ef55. Not hosted here.

### Skill security auditor

Skill · Curated · Security · updated 2026-11-14. Details: https://mikereams.com/ai/skill-security-auditor

Scans a skill before you install it for prompt injection, hidden text, network calls, shell execution and credential reads.

What you get: A pass or fail report on a skill folder, with each risky line flagged.

When to use it: Before installing a skill from a source you do not control.

Ask for it like this: "Audit this skill folder before I install it."

Limits: Static analysis only: it does not run the skill's code, so read the files too.

- Nothing. Its only outside call is a git clone of the repository you name, into a temporary folder.
- Every skill in this library went through it, then a manual read of every file.
- Use it as a first pass, not the only gate: its curl rule misses file uploads sent with curl -F.
- It skips any line the skill under review has marked with the auditor's own ignore comment, so search the skill for that marker yourself.

Source: [alirezarezvani/claude-skills](https://github.com/alirezarezvani/claude-skills/tree/aa8d778/engineering/skills/skill-security-auditor) by Alireza Rezvani (MIT), reviewed at aa8d778. Not hosted here.
