kit/plugins/plan-interview/skills/deep-grill/SKILL.md
Stress-tests plan decisions node-by-node with focused questions. Walks each decision point surfacing assumptions and weak spots. Use when the user asks to deep grill or stress-test a plan.
npx skillsauth add shawn-sandy/agentics deep-grillInstall this skill globally with one command. Works with Claude Code, Cursor, and Windsurf.
3 of 9 scanners reported clean
Some scanners were skipped, did not run, or reported a non-clean status. Review each row below.
Walk each branch of a plan's design tree, asking focused questions at every decision node and exploring the codebase to resolve them.
Before doing any other work, use TodoWrite to create todos for each step of
this session. This gives the user visibility into progress and ensures no step is
skipped.
Create the following todos (all starting with status: "pending"):
Mark each todo status: "completed" as you finish that step.
Use the first match from this priority order:
.md file is
currently open or selected in the IDE (provided via context). If it exists
and its content looks like a plan (contains headings like
## Implementation, ## Plan, ## Steps, ## Context, or ## Summary),
use it..claude/settings.json in the
current project directory. If a "plansDirectory" key exists, glob *.md
files from that path and use the most recently modified file. This takes
precedence over the global config.~/.claude/settings.json. If a "plansDirectory"
key exists, glob *.md files from that path and use the most recently
modified file.Glob on ~/.claude/plans/*.md, sort by
modification time, and select the most recently modified file.If no file can be found via any of these methods, tell the user and stop.
The deep grill is designed for implementation plans. If the resolved file is a
SKILL.md (filename is SKILL.md or frontmatter contains name: and
description: without plan-style headings), inform the user that the deep grill
targets plans, not skill files, and stop.
Announce the file: "Deep grilling plan: path/to/plan.md"
Read the resolved plan file and extract the design tree:
Identify decision nodes: Scan the plan for architectural choices, technology selections, API designs, data models, integration points, file structure decisions, and any other points where an alternative approach exists.
Organize into branches: Group decision nodes by plan section or concern area. Each branch represents a related cluster of decisions.
Present the outline: Show a brief summary of the branches found:
Found N decision branches:
1. [Branch name] — [brief description]
2. [Branch name] — [brief description]
...
Ask the user via AskUserQuestion whether to grill all branches or
select specific ones:
If the user picks specific branches, present the list with numbers and let them choose.
For each selected branch, one at a time:
"Branch N/M: [branch name]"Glob, Grep,
or Read first, then present your finding as the recommended answer.After completing each branch (except the last), use AskUserQuestion to ask:
If the user stops early, proceed directly to Step 4 with the branches examined so far.
Output a structured summary:
## Deep Grill Summary
### Plan
[path/to/plan.md]
### Branches Examined
[N/M branches completed]
### Key Decisions Resolved
- [Branch]: [Decision confirmed] — [rationale or codebase evidence]
### Open Questions Remaining
- [Any branches or sub-questions the user deferred or skipped]
### Recommendations
- [Amendments to the plan based on deep grill findings]
development
Turns a React component into a social card with preview, code, and props table. Builds a static preview and screenshots react-card.html via Playwright. Use when asked to share a React component.
data-ai
Refine-prompt: interviews users and assembles a structured AI prompt using Anthropic best-practice techniques. Use when the user runs /plan-agent:refine-prompt or asks to refine a prompt.
development
Plan review Agent Team. Reviews HTML implementation plans in parallel, synthesizes findings, and applies improvements in place. Use when the user asks to review or improve an implementation plan.
data-ai
Craft-prompt: interviews users and assembles a structured AI prompt using Anthropic best-practice techniques. Use when the user runs /plan-agent:craft-prompt or asks to craft a prompt.