Daily Workflow
This page is the reusable version of the tutorials. Read it after Core Session Commands, Your First Feature, and The Six-Step Loop.
The core idea is simple: Claude Code works best when the team uses a calm, explicit loop instead of treating every request like a giant one-shot prompt.
1. Start sessions so you can resume them later
Use a named session when the work may take more than a few minutes:
claude -n "feature-oauth"Resume recent work with:
claude -c
claude -r "feature-oauth"Good session names save time later. Rename weak names early:
/rename oauth-feature2. Explore before asking Claude to edit
For any non-trivial change, start with read-only understanding:
Read the relevant files and explain the current flow.
Do not edit yet. Tell me what matters and what looks risky.This protects the session from speculative edits and gives you a chance to correct bad assumptions while the context window is still clean.
3. Use the six-step loop for behavior changes
When the task changes behavior, use this rhythm:
- Explore what exists.
- Plan the smallest change that can work.
- RED by creating a failing test, repro, or expected output.
- GREEN by implementing the smallest change that makes the same check pass.
- Review with
/simplifyand, when relevant,/security-review. - Ship only after the evidence is clear.
The loop matters most for features, bug fixes, refactors with risk, and anything security-sensitive.
4. Use the lightweight path for trivial edits
Not every task needs the full loop.
For a typo, copy change, or tiny docs fix, it is fine to be direct:
Fix the typo in the release note heading. Keep the diff to that line only.The rule of thumb is simple: if the overhead of planning and verification is bigger than the change itself, stay light.
5. Review and debug on purpose
Two commands pull a lot of weight in daily work:
/simplify
/debug- Use
/simplifyafter implementation to catch reuse, quality, and clarity issues before you commit. - Use
/debugwhen you have an actual failure, flaky behavior, or a bug report and you want Claude to investigate before proposing random fixes. - Add
/security-reviewfor auth, permissions, secrets, payments, or anything production-sensitive.
6. Context management
Claude Code sessions usually fail from wasted context before they fail from absolute context limits.
Use context management as part of the normal workflow, not as an emergency-only move.
Watch the session before it drifts
Run:
/contextUse it when:
- the session has been open for a while
- Claude has explored several files or branches of thought
- you are about to start a new phase of work
Compact before the session gets noisy
Run:
/compact retain the current plan, the failing verification, the files in play, and the next actionGood compaction prompts name exactly what must survive. Bad compaction prompts say only “summarize this.”
Interrupt bad turns early
Use Ctrl+C when Claude starts going down the wrong path.
That matters because every unnecessary diff, file scan, and speculative idea burns context you could have spent on the actual task.
Prefer targeted reading over broad scanning
When you already know the area, ask for the specific files or folders instead of the whole repo. Broad exploration is useful early, but repeated broad exploration is a context leak.
Fork or resume instead of overloading one thread
Use:
claude -c
claude -r "feature-oauth"
claude -c --fork-session- resume when you want to continue the same thread
- fork when you need to branch the investigation without muddying the original session
Context-management checklist
- check
/contextbefore the session feels broken - compact around 70-80% instead of waiting for the last moment
- interrupt drift with
Ctrl+C - keep prompts specific
- resume or fork instead of forcing unrelated work into one long session
7. Carry work across time and devices
When a session matters, keep it resumable:
/rename checkout-bug
/rc/renamegives the session a useful name/rcmakes it reachable from the Claude app or Claude Code remote viewclaude -candclaude -r "<name>"bring the same work back later
This is more valuable than it sounds. Teams lose a lot of momentum by treating every session like disposable scratch space.
8. Reach for advanced modes later
The advanced controls are powerful, but they are not the starting point:
/fast
/effort high
/batch
/loopUse them only after the basic daily loop is already stable:
/fastand/efforttune speed versus reasoning depth/batchis for genuinely parallel migrations/loopis for recurring checks or routines, not ordinary coding
If you need them, continue with Scaling Workflows. If you do not, ignore them for now.
Daily workflow in one sentence
Start a named session, explore before editing, make verification explicit, review before shipping, and only add more automation after the simple loop already works.