Install
$ agentstack add skill-pekral-cursor-rules-blueprint ✓ scanned · ✓ verified, works with Claude Code, Cursor, and more.
Security review
✓ PassedNo issues found. Passed automated security review. · v0.1.0 How review works →
- ✓ Prompt-injection patterns
- ✓ Secret / credential exfiltration
- ✓ Dangerous shell & filesystem operations
- ✓ Untrusted network calls
- ✓ Known-malicious package signatures
What it can access
- ✓ Network access No
- ✓ Filesystem access No
- ✓ Shell / process execution No
- ✓ Environment & secrets No
- ✓ Dynamic code execution No
From automated source analysis of v0.1.0. “Used” means the capability is present in the source — more access means more to trust, not that it’s unsafe.
Verified badge
Passed review? Show it. Paste this badge into your README, it links to the public security report.
Reliability & compatibility
Declared compatibility
Compatibility is declared by the source manifest. End-to-end runtime verification is coming, see below.
We're building live execution health for every listing: tool-call success rate, median latency, uptime, and last-checked timestamps, measured, not self-reported. It isn't live yet, so we don't show numbers we can't stand behind.
How agent discovery & health will work →About
Blueprint
Constraints
- Apply
@rules/git/general.mdc— branch, commit, and PR conventions below come from it. - Apply
@rules/compound-engineering/general.mdc— the plan is durable memory the next agent reuses, not throwaway prose. - Apply
@rules/laravel/architecture.mdcwhen the project usespekral/arch-app-services, so each step lands in the correct layer. - Plan only. Do not implement, commit, or push any step.
- Output Markdown only. English only.
- Every step must be one pull request: independently reviewable and mergeable on its own.
- Break the objective into 3-12 steps. Fewer means it likely fits one PR; more means the objective is too broad and should be split first.
- Each step must be executable cold by
@skills/resolve-issue/SKILL.mdor a fresh agent with no prior context. - Do not invent dependencies on tooling that may be absent. Degrade gracefully when GitHub CLI (
gh) is missing.
Use when
- An objective clearly exceeds one pull request (a migration, a multi-layer feature, a cross-cutting refactor).
- Work will span multiple sessions or multiple agents and needs a sequenced, resumable plan.
- You need to know which parts can run in parallel and which must be sequential before starting.
Do not use for single-PR tasks (use @skills/analyze-problem/SKILL.md or @skills/resolve-issue/SKILL.md), for splitting an objective into tracker issues without a dependency graph (use @skills/create-issues-from-text/SKILL.md), or when the user wants one high-leverage suggestion (@skills/smartest-project-addition/SKILL.md / @skills/product-capability/SKILL.md).
Execution
Run the five phases in order. Do not register a plan that has not passed Review.
- Research — Establish ground truth before planning.
- Read the actual code, layers, and conventions the objective will touch. Per
@rules/compound-engineering/general.mdc, find the existing part of the system the work extends before inventing a new abstraction. - Walk
git log/git blameover the affected area to learn how it evolved and what was already tried or reverted. - Detect the git baseline: confirm the repo, the default branch (
master), and whetherghis available (gh auth status). Record the result; it drives the PR template in Output. - For an unfamiliar pattern, library, or security-sensitive surface, consult current authoritative references and cite them.
- Design — Decompose the objective into 3-12 one-PR steps.
- Each step is the smallest change that delivers reviewable value and can merge alone.
- Sequence steps so each builds on a merged predecessor; never assume an unmerged step's code.
- Map dependency edges: for every step, list which steps must merge before it. Steps with no shared files and no dependency edge between them may run in parallel.
- Name the layer each step lands in (Action, Service, Repository, etc.) when
@rules/laravel/architecture.mdcapplies.
- Draft — Write the per-step plan (structure in Output).
- Give every step a self-contained context brief: enough background, file paths, and conventions for a fresh agent to execute it without reading the other steps.
- Write concrete, observable exit criteria per step (tests, behavior, verification commands).
- Add the parallelism summary: the dependency graph and which steps may run concurrently.
- Review — Adversarially check the draft against the anti-patterns below before registering. Fix every hit; re-run until clean.
- A "step" that spans multiple PRs, or is too small to justify its own PR.
- A step whose context brief is not self-contained (assumes prior conversation or another step's unmerged work).
- Missing, unobservable, or untestable exit criteria.
- Hidden ordering: two steps marked parallel that touch the same files or share an unstated dependency.
- A new abstraction where an existing part of the system already fit (
@rules/compound-engineering/general.mdc). - Branch / commit / PR wording that violates
@rules/git/general.mdc(wrong type, base branch, period, attribution). - A plan that assumes
ghexists when Research found it absent.
- Register — Persist the approved plan as Markdown in the repo (e.g.
docs/plans//): oneREADME.mdindex plus one file per step, or a single plan file when small. State the path so the next agent picks it up.
Output
Register a plan index plus one entry per step.
Plan index (README.md): objective in one or two sentences; ordered step list; the parallelism summary; git baseline (default branch master, gh present yes/no).
Per step:
- Title —
type(scope): short description, lowercase type/scope, no trailing period (@rules/git/general.mdc). - Branch — feature branch off
master(e.g.feat/). - Context brief — self-contained background, file paths, and conventions; readable cold, with no reference to other steps' unmerged work.
- Tasks — concrete, ordered actions implementing this one PR.
- Dependencies — the step IDs that must merge first;
noneif it can start immediately. - Exit criteria — observable, verifiable conditions that prove the step is done: tests to pass, behavior to confirm, and the exact verification commands to run.
- PR — title in English; body as Markdown linking the issue (
Closes #N) when one exists. Open withghwhen available; otherwise output the title and body for the human to open manually and say so.
Parallelism summary — the dependency graph (e.g. 1 -> 2 -> 4, 3 parallel to 2) and an explicit list of step sets that may run concurrently because they share no files and no dependency edge.
Done when
- The plan has 3-12 steps, each one mergeable pull request.
- Every step has a self-contained cold-start context brief, explicit dependencies, and observable exit criteria.
- The parallelism summary states what can run concurrently and what must be sequential, with no hidden shared-file conflicts.
- The Review pass found no remaining anti-patterns.
- Branch, commit, and PR templates follow
@rules/git/general.mdcand degrade gracefully whenghis absent. - The plan is registered as Markdown in the repo and its path is reported.
Source & license
This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.
- Author: pekral
- Source: pekral/cursor-rules
- License: MIT
- Homepage: https://pekral.cz
Install and usage instructions live in the source repository linked above.
Reviews
No reviews yet, be the first.
Write a review
Versions
- v0.1.0 Imported from the upstream source.