skills

agent
Guvenlik Denetimi
Uyari
Health Uyari
  • No license — Repository has no license file
  • Description — Repository has a description
  • Active repo — Last push 0 days ago
  • Community trust — 18 GitHub stars
Code Uyari
  • Code scan incomplete — No supported source files were scanned during light audit
Permissions Gecti
  • Permissions — No dangerous permissions requested

Bu listing icin henuz AI raporu yok.

SUMMARY

My personal Claude Code skills

README.md

Skills

My personal Claude Code skills, also usable in Codex. Each skill is a folder under skills/ with a SKILL.md that tells the agent when to activate and what to do.

Installation

npx skills@latest add SrdjanCoric/skills

The installer lists every skill in the repo and lets you pick which ones to install. Run it again later to pull updates.

To install a single skill directly:

npx skills@latest add SrdjanCoric/skills --skill write-well

Skills

The main software workflow is talk-it-through, an optional PRD, to-plan, implement-next-task, and sync-main. Implementation includes task review, README updates when needed, final behavior proof, PR creation, and a wait for CI. sync-main merges the verified PR and closes the local task. Each skill also works on its own.

  • talk-it-through: Interviews you about a plan, design, or idea until the decisions are clear. For software work, it inspects the codebase and applies the relevant repository guidelines before saving the agreed decisions.
  • write-a-prd: Turns the current conversation or a decision document into a PRD and saves it to a local file.
  • to-plan: Turns a PRD, decision document, or conversation into the smallest independently verifiable vertical tasks under plans/tasks/. It checks relevant repository guidelines, records direct dependencies, presents the proposed breakdown, and waits for approval before writing files.
  • implement-next-task: Claims one eligible task and implements it on its feature branch. It uses TDD for behavior changes, resolves real uncertainty through talk-it-through, applies repository guidelines, runs autonomous review and remediation, updates the README when needed, proves the final behavior, and opens a CI-green PR after approval.
  • task-review: Reviews a branch against repository standards, the task or spec, correctness, and relevant security risks. It automatically fixes current-diff non-security findings and reruns the review for up to two remediation passes. Every security finding requires a plain-English user decision. Results stay in context instead of being written to a review file.
  • software-repository-guidelines: Applies a shared repository checklist during scoping, planning, implementation, review, or a full final assessment. It loads only the relevant references for ordinary work and requires repository or CI evidence before calling an item complete.
  • create-pr: Commits the current unit of work, pushes the branch, opens or updates its PR, and waits for CI. It invokes diagnose for in-scope CI failures and stops when the current head is ready to merge.
  • sync-main: Treats an explicit invocation as approval to merge a verified, CI-green PR. It synchronizes local main, removes branches whose merged state is proven, changes matching task pointers from [>] to [x], and moves their task files to plans/tasks/done/.
  • write-well: Writes and revises prose in a direct, human voice. It includes a separate adversarial audit that removes common machine-written tells.
  • tdd: Builds features and fixes bugs through observed red-green-refactor cycles. It favors behavior through public interfaces and applies the testing section of the repository guidelines.
  • handoff: Captures one focused slice of the current work into a handoff document that a fresh agent can continue without re-deriving the context.
  • diagnose: Works hard bugs and performance regressions through a disciplined loop: build a feedback loop, reproduce, minimize, hypothesize, instrument, fix, and add a regression test.
  • teach: Turns the current directory into a persistent teaching workspace and teaches one topic over many sessions, one self-contained HTML lesson at a time. Each lesson can collect highlighted questions that shape the next lesson.

Dependencies

Some skills invoke others, shown below with . The installer does not follow these links, so --skill installs only the named skill.

talk-it-through    → software-repository-guidelines
to-plan            → software-repository-guidelines
implement-next-task → software-repository-guidelines, tdd, talk-it-through,
                      task-review, write-well, create-pr
task-review        → software-repository-guidelines, write-well, tdd when remediation needs it
create-pr          → write-well, diagnose
tdd                → software-repository-guidelines
write-a-prd        → write-well
teach              → talk-it-through, write-well
diagnose           → handoff

task-review also uses the environment's code-review capability. It runs a security lens only when the diff touches a relevant trust boundary. In Claude Code it uses /security-review; in Codex it uses $codex-security:security-diff-scan, which requires the Codex Security plugin.

To install implement-next-task and its full transitive dependency chain:

npx skills@latest add SrdjanCoric/skills \
  --skill implement-next-task \
  --skill software-repository-guidelines \
  --skill tdd \
  --skill talk-it-through \
  --skill task-review \
  --skill write-well \
  --skill create-pr \
  --skill diagnose \
  --skill handoff

Run npx skills@latest add SrdjanCoric/skills without --skill to choose from the full list.

The lifecycle states are [ ] ready, [~] in progress, [>] complete with a CI-green PR awaiting merge, and [x] merged into main. Only sync-main changes [>] to [x].

Yorumlar (0)

Sonuc bulunamadi