lsd

Little Shit Done (LSD)

Lightweight planning and execution system. Each step runs in fresh context — the user does /clear between steps. The .task/ directory is the only shared memory between steps.

Commands

Command	What it does	Reads	Writes
/lsd [description]	Gather requirements	—	BRIEF.md
/lsd plan	Create execution plan	BRIEF.md	PLAN.md
/lsd exec <N>	Execute phase N	PLAN.md, PHASE-*.md	PHASE-N.md
/lsd verify	Verify all validations	PLAN.md, PHASE-*.md	VERIFY.md
/lsd status	Show task state & next step	.task/* (presence)	—
/lsd clean	Delete .task/	—	—

Routing

Parse $ARGUMENTS and execute the matching step below. If no argument or the argument doesn't match a command, treat it as the brief step (the argument is the feature description).

Routing rules:

• plan → Step: Plan
• exec <N> or exec → Step: Execute (N required, prompt if missing)
• verify → Step: Verify
• status → Step: Status
• clean → Step: Clean
• Anything else (or empty) → Step: Brief (treat argument as feature description)

---

Step: Brief

Gather what the user wants. If they provided a description, use it as starting point.

Ask only what's missing — skip questions already answered in the description. Cover these:

1. What — What feature or change?
2. Behavior — How should it work?
3. Constraints — Tech preferences, patterns to follow or avoid?
4. Done — How do you verify it works?

Keep it to 3-5 questions max. Be conversational, not an interrogation.

After gathering answers, create .task/BRIEF.md:

# Brief

## Feature
{what they want — be specific}

## Expected Behavior
{how it should work, user-facing description}

## Constraints
{tech choices, patterns, things to avoid — "None" if none}

## Done Criteria
{concrete, verifiable conditions for "done"}

After completing, tell the user:

Brief saved to .task/BRIEF.md
Next: review the brief, then /clear and run /lsd plan

---

Step: Plan

Read the planner instructions from agents/planner.md (path relative to this skill's directory). Then follow those instructions to create the plan.

After completing, tell the user:

Plan saved to .task/PLAN.md
Review the plan. If you want changes, edit .task/PLAN.md directly.
Next: /clear and run /lsd exec 1

---

Step: Execute

Parse the phase number N from the argument (exec 1, exec 2, etc.). If missing, prompt.

Read the executor instructions from agents/executor.md (path relative to this skill's directory). Then follow those instructions to execute phase N.

After completing, tell the user:

Phase N complete. Summary saved to .task/PHASE-N.md
Review the code changes and the summary.
Next: /clear and run /lsd exec {N+1}  (or /lsd verify if this was the last phase)

---

Step: Verify

Read the verifier instructions from agents/verifier.md (path relative to this skill's directory). Then follow those instructions to verify all phases.

After completing, tell the user:

Verification saved to .task/VERIFY.md
If all passed: /clear and run /lsd clean
If failures: fix manually or re-run /lsd exec <N> for the failing phase

---

Step: Status

Check which of these files exist in .task/ (presence only — do not read their contents):

• BRIEF.md
• PLAN.md
• PHASE-*.md (note which numbers)
• VERIFY.md

Print a compact summary (~10-15 lines), then infer and suggest the next command:

State	Next command
No files	/lsd <description>
BRIEF only	/lsd plan
PLAN, no PHASE-1	/lsd exec 1
PHASE-N exists, no PHASE-N+1 (more phases remain per PLAN.md)	/lsd exec N+1
All phases done, no VERIFY	/lsd verify
VERIFY exists	/lsd clean

---

Step: Clean

Delete the entire .task/ directory. Confirm with the user before deleting.

Cleaned up .task/ — all done!