| name | capture |
| description | Captures a rough idea into a refined backlog work item. Use this skill whenever the user says "capture this", "add to backlog", "new work item", "I want to build...", "let's add a feature", "capture idea", or describes something they want to implement next. Also trigger when the user is brainstorming and wants to turn a conversation into a concrete, trackable piece of work. This skill researches what already exists (code, specs, prior research) before drafting, then refines the item interactively with the user.
|
Capture
Turn a rough idea into a refined, actionable backlog item. Every work item is grounded in
what the project already has — the feature spec, existing code, and any prior research.
Workflow
Step 1: Understand the idea
Take whatever the user said — a sentence, a paragraph, a vague wish — and restate it back
in one or two sentences to confirm you understand the intent. If the idea is ambiguous,
ask one focused clarifying question before proceeding.
Step 2: Gather context
Read these files to understand the current state of the project:
- feature.md — the feature specification. What user stories and constraints already exist?
- backlog.md — if it exists, what work items are already captured? Avoid duplicates
and find the next available WI number.
- Source code in src/ — scan the main
.cs files (not obj/bin) to understand what's
already implemented. The idea might overlap with existing functionality.
- Research artifacts — check if any prior research exists (look for files like
research.md, research/ directory, or notes in feature.md itself). Prior research
from the feature-researcher skill may contain domain knowledge, sources, and edge cases
that should inform the work item.
Step 3: Assess research needs
After gathering context, decide whether the idea touches domain concepts that the team
hasn't researched yet. Signs that more research would help:
- The idea mentions a domain term not defined in feature.md (e.g., "tax-loss harvesting",
"rebalancing bands", "minimum trade size")
- The acceptance criteria are hard to write because you don't know the real-world rules
- The idea involves regulatory or financial concepts where getting it wrong matters
If more research is needed, tell the user:
"This touches [concept], which isn't covered in our current research. I'd suggest running
the feature-researcher skill first to gather domain knowledge, then coming back here
to capture the work item. Want to do that?"
Let the user decide. If they say no, do your best with what's available.
Step 4: Draft the work item
Write the work item in this format:
## WI-NNN: [Title]
**Story:** As a [role], I want [capability] so that [benefit]
**Acceptance Criteria:**
- Given [context] When [action] Then [outcome]
- Given [context] When [action] Then [outcome]
**Research:** [links, notes, or "none — consider running feature-researcher"]
**Depends on:** [WI-NNN / "none" / reference to existing user story in feature.md]
**Size:** [S / M / L]
**Status:** todo
Guidelines for the draft:
- Title: short, action-oriented (e.g., "Add transaction cost awareness to rebalance plan")
- Story: identify the real user and real benefit, not just "as a user I want this feature"
- Acceptance criteria: write 2-4 concrete Given/When/Then scenarios. Cover the happy path
and at least one edge case. These should be specific enough that someone could write a test
from them.
- Research: cite any sources found in prior research or feature.md. If no research exists
for this area, note that explicitly.
- Dependencies: reference existing work items or user stories this builds on. Check if
the idea requires something that doesn't exist yet — if so, flag it.
- Size: S = a few hours, M = a day, L = multiple days. Base this on what you saw in the
codebase — if similar things already exist, it's probably smaller.
Step 5: Refine with the user
Present the draft and ask: "How does this look? Anything to add, change, or split?"
Common refinements:
- The user might want to split a large item into smaller ones
- Acceptance criteria might need adjusting based on the user's actual intent
- Dependencies might be wrong or missing
- The user might realize they need research first after all
Iterate until the user confirms the item is ready.
Step 6: Write to backlog
Once confirmed, append the work item to backlog.md in the project root.
-
If backlog.md doesn't exist, create it with a simple header:
# Backlog
Then append the item below.
-
If it exists, read it first to find the last WI number and increment.
After writing, confirm: "Added WI-NNN to backlog.md."
Numbering
Work items use sequential IDs: WI-001, WI-002, etc. If backlog.md already exists,
read it to find the highest number and increment. If it doesn't exist, start at WI-001.
Multiple items in one session
If the user wants to capture several ideas, process them one at a time through the full
workflow. Don't batch-draft — each item deserves its own context-gathering and refinement cycle.