| name | source-command-continue |
| description | Continue incomplete work - find gaps, finish features, fix what's broken |
source-command-continue
Use this skill when the user asks to run the migrated source command continue.
Command Template
Continue
Find what's incomplete. Finish it.
Mandate
- Think, don't checklist. Understand the project first. What is it trying to do? What would "done" look like?
- Delegate workers for parallel research. You synthesize and verify.
- Fix, don't report. Implement solutions directly.
- One pass. No deferrals. Complete each fix fully.
How to Find Incomplete Work
Don't grep for TODO and call it done. Incomplete work hides in:
What's explicitly unfinished — Yes, scan for TODO/FIXME/HACK. But ask: why are they there? What was the person avoiding?
What's implicitly broken — Code that "works" but:
- Fails silently (empty catch blocks, swallowed errors)
- Works only in happy path (no validation, no edge cases)
- Works but confuses users (unclear errors, missing feedback)
- Works but can't be debugged (no logs, no context)
What's missing entirely — Features referenced but not implemented. UI that leads nowhere. Promises in docs that code doesn't deliver.
The Real Test
For each part of the system, ask:
"If I were a user trying to accomplish their goal, where would I get stuck?"
"If this broke at 3am, could someone figure out why?"
"If requirements changed tomorrow, what would be painful to modify?"
"If we had 100x traffic, what would fall over first?"
These questions reveal incompleteness that checklists miss.
Execution
-
Understand the project — Read README, main entry points, core flows. What is this thing supposed to do?
-
Walk the critical paths — Trace actual user journeys through code. Where does the path get uncertain, error-prone, or incomplete?
-
Find the gaps — Not just TODOs, but:
- Dead ends (code that starts something but doesn't finish)
- Weak spots (minimal implementation that will break under pressure)
- Missing pieces (what's referenced but doesn't exist)
-
Prioritize by impact — What blocks users? What causes data loss? What makes debugging impossible? Fix those first.
-
Fix completely — Don't patch. Understand root cause. Implement proper solution. Test it works.
Skills
Domain knowledge auto-activates when relevant. Available skills:
auth, account-security, privacy, billing, pricing, ledger, security, trust-safety, uiux, seo, pwa, performance, i18n, database, data-architecture, storage, observability, operability, delivery, growth, referral, support, admin, discovery, code-quality
Loop
After fixing: "Did my fixes introduce new gaps? Did fixing X reveal Y was also broken?"
If yes → run /continue again. If no Critical/High issues remain → done.
Output
## What I Found
[Describe the gaps discovered — not a checklist, but an understanding of what's incomplete and why]
## What I Fixed
- [Description of fix and why it matters]
## What Remains
- [Issues that need human decision or are blocked]
## Next
[/continue again | done]