Back to search
agentic-dev-workflow
MarketplaceWorks with Claude CodeGitHub
Two Claude Code skills for agentic development, with an eval harness
Safety scan
A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.
Badge for your README
Markdown
[](/listings/fuggerstadt1907--agentic-dev-workflow)HTML
<a href="/listings/fuggerstadt1907--agentic-dev-workflow"><img src="/api/listings/fuggerstadt1907--agentic-dev-workflow/badge.svg" alt="findsafeskills safety rating" /></a>Install
Install instructions from the README
As written by the repo's author. findsafeskills has not verified these steps.
Inside Claude Code:
/plugin marketplace add fuggerstadt1907/agentic-dev-workflow /plugin install agentic-dev-workflow@agentic-dev-workflow
Other ways (copying the skills, one project only) and updating: [install guide](docs/install.md).
What was scanned
The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.
.claude-plugin/marketplace.json · Manifest · 17 lines
- {
- "name": "agentic-dev-workflow",
- "owner": {
- "name": "Alessandro Orlandi",
- "url": "https://github.com/fuggerstadt1907"
- },
- "metadata": {
- "description": "Marketplace for the agentic-dev-workflow plugin: grill-ticket and implement-ticket."
- },
- "plugins": [
- {
- "name": "agentic-dev-workflow",
- "source": "./",
- "description": "grill-ticket and implement-ticket: from ticket to requirements to reviewed, committed stages."
- }
- ]
- }
README.md · README · 100 lines
- # agentic-dev-workflow
- Two Claude Code skills that take a ticket to working, reviewed code. `grill-ticket` checks the
- facts itself, asks you only for decisions, and writes a requirements document with stages.
- `implement-ticket` builds one stage at a time, has a fresh subagent review it, and commits.
- > Evals in progress – results pending.
- ```mermaid
- flowchart LR
- B[Brief / ticket] --> G["/grill-ticket"] --> R[requirements.md]
- R --> P
- subgraph I["/implement-ticket, once per stage"]
- P[Plan] --> S[Thin slice] --> V[Fresh-context review]
- V --> F[Fix] --> C[Commit]
- end
- C -. next stage .-> P
- C --> W[Worklog]
- ```
- Brief → /grill-ticket → requirements → /implement-ticket per stage (plan → thin slice →
- fresh-context review → fix → commit, repeat) → worklog
- ## Install
- Inside Claude Code:
- ```text
- /plugin marketplace add fuggerstadt1907/agentic-dev-workflow
- /plugin install agentic-dev-workflow@agentic-dev-workflow
- ```
- Other ways (copying the skills, one project only) and updating: [install guide](docs/install.md).
- ## Usage
- ```text
- /grill-ticket #123 → docs/…/requirements.md (Status: ready)
- /implement-ticket docs/…/requirements.md → stage 1 built, reviewed, committed
- /clear
- /implement-ticket docs/…/requirements.md → stage 2 …
- ```
- The short form (`/implement-ticket`) works as long as no other installed skill has the same name.
- The namespaced form always works: `/agentic-dev-workflow:grill-ticket`,
- `/agentic-dev-workflow:implement-ticket`.
- The skills are written in English and work in the language of the project they run in. Each step
- is explained on the [workflow page](docs/workflow.md). Templates for a project's
- [`CLAUDE.md`](templates/CLAUDE.md) and [`WORKLOG.md`](templates/WORKLOG.md) are in `templates/`.
- The goal is that you understand the workflow in 2 minutes and install it in 5. Nobody has measured
- that yet.
- ## Prevention and detection
- Some rules in `implement-ticket` prevent defects: a security-relevant stage gets acceptance
- criteria written from an attacker's view before building, and every test that asserts something
- is absent gets a positive check that what it looks for is really in the input. The review eval does not measure
- these rules.
- Detection is the fresh-context review with
- [`REVIEW.md`](skills/implement-ticket/REVIEW.md). The review eval measures it against a
- plain-prompt baseline, using planted defects and clean diffs.
- ## Why this exists
- Coding agents write code fast. They also build on unclear requirements and fill the gaps with
- assumptions; the agent that wrote the code reviews it and keeps its blind spots; and nobody
- measures whether a prompt or skill helps. These skills come out of a real project. The evals, still
- in progress, compare each skill with a plain prompt and report where the skill does not win, too.
- This repo is built with its own skills. Every stage is recorded in the
- [worklog](docs/WORKLOG.md), and the decisions are in the [requirements](docs/requirements.md).
- ## Credits
- These ideas come from Matt Pocock's [skills](https://github.com/mattpocock/skills):
- - [`grilling`](https://github.com/mattpocock/skills/blob/abeeaf95660ba9e8b7c753eb38f441ca707a8823/skills/productivity/grilling/SKILL.md):
- the design tree worked in rounds (`grill-ticket` calls it the decision tree), the frontier of questions whose prerequisites are settled, and
- the split between facts the agent looks up and decisions the user makes. Used in `grill-ticket`
- and, as the split between decisions the agent may settle and decisions for the user, in
- `implement-ticket`.
- - [`tdd`](https://github.com/mattpocock/skills/blob/abeeaf95660ba9e8b7c753eb38f441ca707a8823/skills/engineering/tdd/SKILL.md):
- tests at agreed seams, each with a one-line note on what it catches and what it misses, and red
- before green. Used in `implement-ticket`'s plan and build steps.
- - [`to-tickets`](https://github.com/mattpocock/skills/blob/abeeaf95660ba9e8b7c753eb38f441ca707a8823/skills/engineering/to-tickets/SKILL.md):
- tracer bullets, narrow slices that cut through every layer. Used in `implement-ticket`'s build
- step.
- No text is copied from these skills.
- ## Contributing
- See [CONTRIBUTING.md](CONTRIBUTING.md).
- ## Licence
- [MIT](LICENSE)
Ratings and comments
Sign in to rate this skill.
Sign inFree to use. If it’s helped, pass it on.