From devproc
Begin implementing a pending feature — move to In Progress and create a plan file
How this skill is triggered — by the user, by Claude, or both
Slash command
/devproc:feature-startThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Start implementing a feature.
Start implementing a feature.
Before proceeding, check that CLAUDE.md contains a Feature Model section. If it doesn't, tell the user to run /feature-init first and stop.
The user has identified the feature as: $ARGUMENTS
Steps:
Read FEATURES.md and identify the feature to start. If $ARGUMENTS is blank and there is exactly one entry in ## Pending, use that. If ambiguous, ask the user to clarify.
Move the feature entry from ## Pending to ## In progress in FEATURES.md. Update the In Progress entry to reference the plan file: add a line like Sub-task detail in [plans/<slug>.md](plans/<slug>.md).
Decide whether to use deep planning mode:
/plan if: the feature is complex (multiple systems touched, significant unknowns, non-trivial architecture decisions), OR the user has asked for deeper thinking or planning (e.g. "use plan mode" or "this is a complex change")./plan for straightforward features where the sub-tasks are obvious from the description./plan: invoke it now before proceeding. The plan output will inform the sub-task breakdown.Before creating the plan file, do enough research to produce a realistic sub-task breakdown:
Create plans/<slug>.md with this structure:
# <Feature title> — Feature Plan
## Handoff
**Last updated:** YYYY-MM-DD
**Session summary:** Feature plan created. Implementation not yet started.
**Sub-task in progress:** None
**First action next session:** Begin Sub-task 1
**Open questions / decisions pending:** None
**Dead ends to avoid:** None
## Design
<The output of the planning process goes here. For a simple feature this may be
a short paragraph. For a complex feature it may be several pages covering
architectural decisions, data layouts, component interactions, and open
questions resolved during planning. This section is the written record of
what was decided and why — enough detail that the user can review it inline
and provide corrections before implementation begins.>
## Sub-tasks
1. **<Sub-task name>** — <brief description of what success looks like>
2. **<Sub-task name>** — ...
...
**▶ NEXT:** Sub-task 1
> Run `/feature-checkpoint` after each sub-task completes.
Keep sub-task descriptions to one line. Implementation detail goes in NOTES.md as you discover it, not here. The Design section is the exception: it should capture the key decisions and rationale from the planning process.
Provides behavioral guidelines to reduce common LLM coding mistakes, focusing on simplicity, surgical changes, assumption surfacing, and verifiable success criteria.
Searches, retrieves, and installs Agent Skills from prompts.chat registry using MCP tools like search_skills and get_skill. Activates for finding skills, browsing catalogs, or extending Claude.
npx claudepluginhub plwhite/claudeplugins --plugin devproc