Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Manage structured communication files (PLAN.md, TODO.md, SPEC.md, etc.) to coordinate work across agents and sessions. Use when starting complex tasks, delegating to subagents, or maintaining project state.
.claude/skills/marco-souza-project-files/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-02 | ✗→✓ | ▲ Improved | 118% | 0% |
| case-01 | ✗→✓ | ▲ Improved | 20% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 79% | 0% |
| case-12 | ✗→✓ | ▲ Improved | 95% | 0% |
| case-15 | ✗→✓ | ▲ Improved | 262% | 0% |
Standard files for coordinating work across agents and sessions. These files serve as the single source of truth for project state.
The 3 core files for agent coordination:
markdown# Plan: <Project> ## Goal One-line objective. ## Strategy High-level approach. ## Phases ### Phase 1: <Name> [IN_PROGRESS] - Objective: What this achieves - Success criteria: How we know it's done
markdown# TODO ## Current Sprint ### In Progress - [ ] agent: Task description ### Ready - [ ] agent: Task description ### Blocked - [ ] agent: Task (blocked by: X)
markdown# Session: YYYY-MM-DD HH:MM ## Context What we were working on. ## Current State - PLAN phase: X, TODO items: Y, Blockers: None ## Next Actions 1. Complete X (assigned to: agent) 2. Start Y
| File | Purpose | Updated By | Read By | |------|---------|------------|---------| | PLAN.md | High-level strategy and approach | Lead agent | All agents | | TODO.md | Current tasks and queue | Working agents | All agents | | SPEC.md | Technical requirements and design | Architect/planner | Implementation agents | | DECISIONS.md | Architecture decisions (ADRs) | Any agent | All future agents | | SESSION.md | Session context and handoff | Current session | Next session | | STATUS.md | Current project state | Any agent | Status checks | | CHANGELOG.md | Completed work log | Working agents | Review agents |
Strategic direction. Updated when approach changes.
markdown# Plan: <Feature/Project Name> ## Goal One-line objective. ## Strategy High-level approach (2-3 paragraphs). ## Phases ### Phase 1: <Name> [IN_PROGRESS] - Objective: What this phase achieves - Success criteria: How we know it's done ### Phase 2: <Name> [PENDING] - Objective: ... ## Constraints - Must use X technology - Must maintain Y compatibility ## Risks - Risk: Mitigation strategy
State markers: [PENDING], [IN_PROGRESS], [COMPLETED], [BLOCKED]
Active work queue. Updated continuously.
markdown# TODO ## Current Sprint ### In Progress - [ ] <agent-name>: Task description (blocked by: X, ETA: Y) ### Ready - [ ] <agent-name>: Task description (depends on: X) - [ ] <agent-name>: Task description ### Blocked - [ ] <agent-name>: Task description (blocked by: X, reason: Y) ## Backlog - [ ] Future task ## Completed (Last 5) - [x] <agent-name>: Task description (completed: DATE)
Assignment format: <agent-name>: <task> (e.g., architect: Design auth API)
Technical specification. Updated during planning.
markdown# SPEC: <Component/Feature> ## Overview What this component does. ## Interface ### API/Exports
function doThing(input: Input): Output
### Typesinterface Input { ... } interface Output { ... }
## Behavior
- Given X, should Y
- Error cases: Z
## Dependencies
- Requires: module-a
- Used by: module-b
## Open Questions
- [ ] Question to resolveArchitecture Decision Records. Append-only.
markdown# Decisions ## ADR-001: <Title> **Date:** YYYY-MM-DD **Status:** Proposed | Accepted | Deprecated | Superseded by ADR-XXX ### Context What problem were we solving? ### Decision What did we decide? ### Consequences Positive: - Benefit 1 Negative: - Trade-off 1
ADR numbering: Sequential (ADR-001, ADR-002...)
Session handoff. Updated at end of session.
markdown# Session: YYYY-MM-DD HH:MM ## Context What we were working on and why. ## Current State - PLAN phase: X - TODO items in progress: Y, Z - Blockers: None | X ## Files Modified - `src/auth/login.ts` - Added validation - `tests/auth.test.ts` - Added tests ## Next Actions 1. Complete X (assigned to: agent-name) 2. Start Y (ready to go) ## Notes for Next Session - Watch out for: potential issue - Remember to: check Z
Quick status dashboard. Updated as needed.
markdown# Status **Last Updated:** YYYY-MM-DD HH:MM **Current Phase:** Implementation **Overall Health:** 🟢 On Track | 🟡 At Risk | 🔴 Blocked ## Progress - [x] Planning complete - [x] Design approved - [ ] Implementation (60%) - [ ] Testing - [ ] Deployment ## Blockers None. ## Recent Changes - Change 1 - Change 2
Completed work history. Append-only.
markdown# Changelog ## [Unreleased] ### Added - Feature X implemented by @agent-name - API endpoint Y added ### Fixed - Bug Z resolved ## [DATE] - Release Name ### Added - Initial implementation
Follows Keep a Changelog format.
bash# 1. Create PLAN.md with high-level strategy cat > PLAN.md << 'EOF' # Plan: User Authentication ## Goal Implement secure JWT-based authentication. ## Strategy ... EOF # 2. Create TODO.md with initial tasks cat > TODO.md << 'EOF' # TODO ## Current Sprint ### Ready - [ ] architect: Design auth API (SPEC.md) - [ ] security: Review threat model EOF # 3. Create initial SPEC.md template cat > SPEC.md << 'EOF' # SPEC: Authentication API ## Overview ... EOF
When using spawn-subagents or mixture-of-experts:
bash# Read context files before spawning PLAN=$(cat PLAN.md) TODO=$(cat TODO.md) SPEC=$(cat SPEC.md) # Spawn expert with full context tmux send-keys -t subagent-architect \ "pi -p 'Review PLAN.md, TODO.md, and SPEC.md. Then: $TASK' ..." C-m
At end of session:
bash# 1. Update TODO.md - mark completed, move in-progress to ready # 2. Update SESSION.md with current state # 3. Update CHANGELOG.md with completed work cat >> CHANGELOG.md << 'EOF' ### Added - Implemented user login flow (SESSION.md for details) EOF
Use files to share context between experts:
bash# Pre-populate files for experts to read for expert in architect security performance; do # Each expert reads PLAN/TODO/SPEC, writes to DECISIONS tmux send-keys -t "moe-$expert" \ "cat PLAN.md SPEC.md | pi -p 'Review and add decisions to DECISIONS.md' ..." C-m done
[IN_PROGRESS], [BLOCKED], [DONE]PLAN.md → Created at project start, updated when strategy shifts
TODO.md → Created with first tasks, updated continuously
SPEC.md → Created during design phase, refined during implementation
DECISIONS.md → Created on first ADR, appended forever
SESSION.md → Created at session end, archived after next session starts
STATUS.md → Created when needed, kept current
CHANGELOG.md → Created at project start, appended foreverbash# Check current status cat STATUS.md # See what's in progress grep -A 5 "In Progress" TODO.md # View recent decisions tail -50 DECISIONS.md # Find previous session context cat SESSION.md # See what changed recently tail -30 CHANGELOG.md
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-02 | fail→pass | 8,036 | 5,889 | -27% | 1 | 1 | 0% | 1,586 | 3,452 | +118% | 0 | 0 | — |
case-01 | fail→pass | 21,552 | 11,915 | -45% | 1 | 1 | 0% | 3,648 | 4,389 | +20% | 0 | 0 | — |
case-03 | fail→pass | 12,244 | 9,237 | -25% | 1 | 1 | 0% | 2,176 | 3,896 | +79% | 0 | 0 | — |
case-04 | fail→fail | 11,926 | 10,152 | -15% | 1 | 1 | 0% | 2,324 | 4,318 | +86% | 0 | 0 | — |
case-05 | fail→fail | 11,846 | 7,610 | -36% | 1 | 1 | 0% | 2,344 | 3,652 | +56% | 0 | 0 | — |
case-06 | fail→fail | 11,414 | 11,038 | -3% | 1 | 1 | 0% | 2,474 | 4,393 | +78% | 0 | 0 | — |
case-07 | fail→fail | 9,011 | 7,524 | -17% | 1 | 1 | 0% | 1,595 | 3,566 | +124% | 0 | 0 | — |
case-08 | fail→fail | 5,847 | 3,947 | -32% | 1 | 1 | 0% | 1,151 | 2,977 | +159% | 0 | 0 | — |
case-09 | pass→pass | 7,818 | 6,532 | -16% | 1 | 1 | 0% | 1,788 | 3,662 | +105% | 0 | 0 | — |
case-10 | pass→pass | 5,723 | 5,869 | +3% | 1 | 1 | 0% | 1,195 | 3,454 | +189% | 0 | 0 | — |
case-11 | pass→pass | 5,725 | 4,091 | -29% | 1 | 1 | 0% | 1,112 | 2,933 | +164% | 0 | 0 | — |
case-12 | fail→pass | 14,017 | 13,928 | -1% | 1 | 1 | 0% | 2,546 | 4,977 | +95% | 0 | 0 | — |
case-13 | pass→pass | 13,084 | 7,329 | -44% | 1 | 1 | 0% | 1,976 | 3,643 | +84% | 0 | 0 | — |
case-14 | fail→fail | 16,601 | 7,355 | -56% | 1 | 1 | 0% | 3,683 | 3,728 | +1% | 0 | 0 | — |
case-15 | fail→pass | 4,533 | 2,541 | -44% | 1 | 1 | 0% | 735 | 2,662 | +262% | 0 | 0 | — |
case-16 | fail→pass | 9,319 | 9,018 | -3% | 1 | 1 | 0% | 1,917 | 4,164 | +117% | 0 | 0 | — |
case-17 | fail→pass | 6,112 | 2,167 | -65% | 1 | 1 | 0% | 1,181 | 2,630 | +123% | 0 | 0 | — |
case-18 | fail→pass | 11,284 | 3,009 | -73% | 1 | 1 | 0% | 2,030 | 2,868 | +41% | 0 | 0 | — |
case-19 | fail→pass | 10,746 | 6,522 | -39% | 1 | 1 | 0% | 2,020 | 3,534 | +75% | 0 | 0 | — |
case-20 | fail→pass | 10,415 | 7,149 | -31% | 1 | 1 | 0% | 1,967 | 3,486 | +77% | 0 | 0 | — |
case-21 | fail→pass | 7,751 | 4,419 | -43% | 1 | 1 | 0% | 1,449 | 3,147 | +117% | 0 | 0 | — |
case-22 | fail→pass | 11,078 | 5,360 | -52% | 1 | 1 | 0% | 2,109 | 3,255 | +54% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted. The headline lift of +55 percentage points is the difference between those two pass rates over the 22 comparable cases.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.