Skip to content
Agentmart
Submit a tool

align

Use before a substantial change to agree scope, design, acceptance criteria, and verification in a signed spec.

Skill by Camgineer

IndexedMIT
Write a review

Install

Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.

npx agentmarket install skill/camgineer-align

Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.

Get the weekly diff

The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.

We use your email only for this newsletter. Privacy

About

Use before a substantial change to agree scope, design, acceptance criteria, and verification in a signed spec.

Use cases: design

No reviews yet. Installed it? Reviews come from verified installers only. How reviews work

Permissions

What this skill can do and what it asks for, from its SKILL.md. Checked Oct 25, 2026.

Tools
No tools found
Secrets
None asked for
Install scripts
None found

Network hosts it talks to 1

  • codeload.github.com · found in install command

Files it touches

  • ~/.claude/skills/align

What's inside

Skill name
align
Description
Use before a substantial change to agree scope, design, acceptance criteria, and verification in a signed spec.
License
MIT
Metadata
source: Adapts to-spec, to-tickets, and wayfinder from mattpocock/skills, https://github.com/mattpocock/skills
Instructions
46 lines in SKILL.md · view on GitHub

Files (6)

  • LICENSE
  • MAP.md
  • SKILL.md
  • SPEC-TEMPLATE.md
  • openai.yaml · agents/
  • origin.json

SKILL.md (first lines)

# Align

Cut through the fog of war before the build. The operator does the critical thinking once, at the start. You do every lookup, draft, and sketch so their time goes only to decisions. End with a **signed spec** that an implementer can use without redoing the decisions.

## 1. Ground

Before any design question, check whether the operator can use what exists today. Inspect the existing command, installation, or output that the request builds on. When a read-only check is available, run it within the operator's authorization. In your first reply after align is invoked, report what you checked and what you found.

Report whether existing use works, is blocked, or has no existing system to build on. For a block, name the thinnest existing recovery path and its observable check. Take an authorized recovery path before design questions. A code change needs an agreed scope and verification first, even when it is small. Record the result in the spec, then design the remaining work.

When recovery needs a decision or permission, ask only for that decision and keep its check pending. If recovery is unavailable or declined, record the limitation and continue design. The requested feature is not itself an existing-use blocker.

Run the **how** skill over every subsystem the change touches. Read `GLOSSARY.md` and the ADRs in that area. Done when you can name each module the change touches and the tests that already cover it.

## 2. Tier

State a tier with a one-line reason drawn from blast radius and reversibility, at the top of the first Grill round. It is a default the operator overrides with one word, not a question.

| Tier | When | Spec sections |
| --- | --- | --- |
| Inline | A typo, a one-line fix with an obvious check | None, except the short spec in step 1 for an unsigned slice. State the definition of done in one line and proceed on the operator's yes. |
| Light | Small, reversible, one surface | Problem, Definition of done, Acceptance criteria, Verification plan, Out of scope |
| Full | A new module, a public interface, a one-way door, or work the operator steps away from | Every section of [SPEC-TEMPLATE.md](SPEC-TEMPLATE.md) |
| Map | Too big or too foggy for one session | [MAP.md](MAP.md), which ends in a Full spec |

## 3. Grill

Run the **grill-with-docs** skill with the tier's spec sections as the design tree, in template order. Problem and definition of done are the root. Every later section hangs off them. Follow its document, reply, and answer-settlement method for every round.

- **Designs.** In the Full tier, run the **architect** skill with checkpoint, and stop at its checkpoint: no implementation and no commit. Its design and the alternative shapes become the options for the high- and low-level design questions.
- **Acceptance criteria.** Each one is behavioral, observable, and checkable on its own: "running `export --tag work` writes only rows tagged work", never "export works". Name interfaces and types, not file paths. Give each an ID: `AC-1`, `AC-2`.
- **Verification.** Testing is a product decision, so put it to the operator. For each criterion agree the public seam (per the [test-behavior-not-implementation](../principles/test-behavior-not-implementation.md) reference), the test, any eval, the live check on the real surface, and the evidence it leaves in the PR. Prefer existing seams and the highest one available; the ideal is one. Write no test at a seam the operator has not confirmed. Mark the criteria that must always hold; they become regression tests.
- **Slices.** Cut the work into vertical tracer bullets, each a narrow, complete path through every layer, demoable alone, sized for one fresh context, with prefactoring first. Order any slices that restore existing use before the feature's slices, with only the prefactoring each restoration path needs. Each slice is one atomic PR in a stack. Atomic means one purpose a reviewer can state in a sentence, not the smallest diff. A new type, field, helper, or endpoint ships in the slice with its first caller, so no slice lands code nothing calls. Split by purpose, never by layer or line count. Each commit inside a slice is one coherent step, and fixup commits are fine. Sequence a wide mechanical refactor as expand with the first migrated caller, migrate the rest in batches, then contract. Ask whether the granularity is right, whether each blocking edge truly gates, and what to merge or split.
- **Autonomy envelope.** State the template's tripwires and the low-level calls the loop will make alone as a default in the last round, with a one-word override. When a decision covers a class, such as un-exporting every symbol only tests import, authorize the class and list the known instances as examples.

## 4. Draft

Fill [SPEC-TEMPLATE.md](SPEC-TEMPLATE.md) for the tier, using the glossary's terms. Keep the draft in an available shared document or an untracked working file. When the change adds a setting that names a command to run, add one line naming where that setting may come from, and allow only the operator's own config. An operator's waiver of a safeguard covers their own tools, not content a repository ships.

Check the spec against the current source and every acceptance criterion. List the questions an implementer would need answered. Fill in facts from evidence and put product decisions to a Grill round. Repeat until no question remains.

## 5. Sign

Show the whole spec and ask for sign-off. Sign-off is the operator's explicit word, such as "aligned". Silence and partial answers keep the spec unsigned. Record unresolved concerns and any explicit override in the spec. End with the signed spec; it does not authorize implementation beyond the user's existing grant.

From Camgineer/cstack, MIT.

More from Camgineer/cstack

Signals

Stars
0
Forks
0
Open issues
0
Last commit (this folder)
Oct 12, 2026 (17 days ago)
Last release
Oct 16, 2026 (13 days ago)

From GitHub (fetched Oct 24, 2026).

Details

License
MIT
First seen
Oct 24, 2026
Last checked
Oct 24, 2026

More from Camgineer

Cite this page

align skill. Agent Mart. https://agentmart.dev/skills/camgineer-align (last updated 2026-10-24)