Skip to main content

PILOT Execution Skill

Overview

The PILOT Execution Skill orchestrates complete project execution lifecycle using:

  • Track-based nomenclature (CODITECT-STD-001)
  • MoE Task Execution methodology (5-step process)
  • Agent discovery and invocation
  • Context preservation and session logging
  • Automated document classification

When to Use This Skill

Use this skill when:

  1. Starting a development session - Need full project orientation
  2. Executing PILOT plan tasks - Working on Track A-G tasks
  3. Managing session context - Need context preservation and logging
  4. Orchestrating multi-agent work - Coordinating multiple agents on related tasks
  5. Creating checkpoints - After completing major milestones

Skill Invocation

Via Command

/pilot              # Full workflow
/pilot --status # Status only
/pilot --track A # Specific track
/pilot --task A.1.1 # Specific task

Via Skill Reference

Use the pilot-execution skill to orchestrate Track A backend tasks.

Via Agent

/agent pilot-orchestrator "Execute Track A.1 Heartbeat System tasks"

The 4-Phase Process

Phase 1: Session Start

# Orientation
/orient

# Context loading
/cxq --recent 200
/cxq --decisions --limit 20

# Agent discovery
/which <next-task>

# PILOT plan review
Read("PILOT-PARALLEL-EXECUTION-PLAN.md")

Phase 2: Task Selection

# Identify next track using priority order:
# 1. Blocking tracks
# 2. P0 Critical Path
# 3. Active tracks
# 4. Unblocked tasks

# Create invocation with Track ID
invocation = Task(
subagent_type="backend-development",
prompt="Implement Track A.1.1: Background heartbeat thread..."
)

# Update tasklist with invocation
# Add to task item in PILOT plan

Phase 3: Task Execution

# Execute agent invocation
result = Task(subagent_type="backend-development", prompt="...")

# OR use CODITECT agent
/agent backend-development "Implement Track A.1.1..."

# Verify completion
# - Code compiles
# - Tests pass
# - No security issues

# Update PILOT plan
# Mark [x] complete with details

# Classify new documents
/classify path/to/new/file.md

Phase 4: Session Management

# Log to session log
/session-log "Track A.1.1 Complete" \
--issue "..." \
--fix "..." \
--files "..."

# Capture context
/cx

# Create checkpoint (if milestone)
/checkpoint "Track A.1 Complete"

Track Nomenclature (CODITECT-STD-001)

TrackDomainAgentRepository
ABackend APIbackend-developmentcoditect-cloud-infra/backend
BFrontend UIfrontend-react-typescript-expertcoditect-cloud-frontend
CDevOpsdevops-engineerInfrastructure/K8s
DSecuritysecurity-specialistCross-cutting
ETestingtesting-specialistTest suites
FDocscodi-documentation-writerDocs sites
GDMSprompt-analyzer-specialistDocument Management

Task ID Format

Track <LETTER>.<MAJOR>.<MINOR>[.<SUB>]

Examples:
- Track A.1 → Backend: Heartbeat System
- Track A.1.1 → Backend: Heartbeat: Background thread
- Track B.3.4.2 → Frontend: Commerce UI: Entitlements: Access badges

Integration Points

Commands Used

CommandPhasePurpose
/orient1Session orientation
/cxq1Context query
/which2Agent discovery
/agent3CODITECT agent invocation
/classify3Document classification
/session-log4Session logging
/cx4Context capture
/checkpoint4Milestone checkpoint

Skills Used

SkillPurpose
moe-task-execution5-step MoE methodology
session-loggingSession log management
context-preservationContext capture and query

Hooks Triggered

HookEventAction
post-session-log-update.pySession log updateAuto-classify
session-retrospective.pySession endCapture learnings

Success Criteria

CriteriaMeasurement
Task CompletionTask marked [x] in PILOT plan
VerificationCode compiles, tests pass
DocumentationSession log updated
ClassificationNew docs classified (moe_confidence > 0.85)
Context Preserved/cx executed

Error Handling

Task Failure

1. Log failure to session log
2. Mark task as failed in PILOT plan
3. Analyze cause (agent, code, infrastructure)
4. Present recovery options
5. Retry or escalate

Agent Timeout

1. Check partial output
2. Create continuation prompt
3. Retry with extended timeout
4. Log timeout event

Context Issues

1. Warn about limited context
2. Fall back to file-based context
3. Suggest /cx --rebuild

Anti-Patterns to Avoid

Anti-PatternCorrect Approach
Skip orientationAlways run Phase 1
Execute without /whichAlways get agent recommendation
Skip session loggingAlways log significant tasks
Ignore classificationAlways classify new documents
Batch completion updatesUpdate PILOT plan immediately

Examples

Full Session

# Start
/pilot

# Execute recommended tasks
# ... agent work ...

# End
/cx
/session-log --view

Track-Specific Work

# Focus on security
/pilot --track D

# Execute D.2 tasks
# ... security hardening ...

# Log completion
/session-log "Track D.2 P1 Security Complete"

Single Task

# Execute specific task
/pilot --task A.1.1

# Verify and log
/session-log "Track A.1.1 Heartbeat thread implemented"

Success Output

When successful, this skill MUST output:

✅ SKILL COMPLETE: pilot-execution

Completed:
- [x] Session oriented with /orient
- [x] Context loaded from last 200 messages
- [x] Agent discovery performed for next tasks
- [x] Task invocations created and added to PILOT plan
- [x] Tasks executed with proper tracking
- [x] PILOT plan updated with completion status
- [x] New documents classified with /classify
- [x] Session logged and context captured

Current Track Status:
- Track A (Backend): 3 tasks completed this session
- Track B (Frontend): 1 task completed this session
- Track F (Docs): 2 documents classified

Outputs:
- Updated PILOT-PARALLEL-EXECUTION-PLAN.md with [x] completed tasks
- Session log entry in docs/session-logs/SESSION-LOG-YYYY-MM-DD.md
- Context database updated via /cx
- Classified documents with moe_confidence > 0.85

Completion Checklist

Before marking this skill as complete, verify:

  • /orient executed at session start
  • /cxq --recent 200 loaded context
  • /which consulted for agent recommendations
  • Task invocations added to PILOT plan before execution
  • All executed tasks marked [x] in PILOT plan
  • New/modified documents classified with /classify
  • Session log updated with issues/fixes/files
  • /cx executed to capture session context
  • No blocking issues left unresolved
  • Track progress reflects actual completion

Failure Indicators

This skill has FAILED if:

  • ❌ Session started without running /orient
  • ❌ Executed tasks without consulting /which first
  • ❌ Tasks completed but PILOT plan not updated
  • ❌ New documents created but not classified
  • ❌ Session log missing or incomplete
  • ❌ Context not captured with /cx
  • ❌ Task invocations not added to PILOT plan
  • ❌ Wrong agent selected for task type
  • ❌ Track IDs incorrect or inconsistent
  • ❌ Blocking issues not escalated

When NOT to Use

Do NOT use this skill when:

  • Working outside the PILOT project (use general task execution)
  • Only need to view PILOT plan status (use /pilot --status)
  • Running ad-hoc experiments (no need for full orchestration)
  • Simple bug fixes not in PILOT plan (use direct agent invocation)
  • Research tasks without deliverables (use research skills)
  • Quick documentation updates (use /agent codi-documentation-writer)
  • Emergency hotfixes (bypass full workflow for speed)

Use alternative skills:

  • moe-task-execution - General MoE task execution without PILOT context
  • session-logging - Session log management only
  • context-preservation - Context capture without full orchestration
  • agent-discovery - Agent selection without execution

Anti-Patterns (Avoid)

Anti-PatternProblemSolution
Skip /orientMissing context, wrong prioritiesAlways run Phase 1 orientation
Execute without /whichWrong agent selectedAlways consult agent discovery first
Batch PILOT updatesPlan becomes staleUpdate immediately after task completion
Skip classificationUntracked documentsAlways /classify new/modified docs
No session loggingLost context between sessionsAlways log significant work
Incomplete task IDsTrack confusionUse full Track.Section.Task format
Work across tracksContext switching overheadFocus on one track per session
Skip verificationFalse completion claimsAlways verify code compiles and tests pass

Principles

This skill embodies these CODITECT principles:

  • #1 Separation of Concerns - Track-based organization separates domains
  • #2 First Principles - MoE methodology ensures proper agent selection
  • #4 Self-Provisioning - Auto-discovery of agents and context
  • #5 Eliminate Ambiguity - Clear task IDs and status tracking
  • #6 Clear, Understandable, Explainable - Session logs document decisions
  • #7 Inform, Don't Ask - Automated context loading and classification
  • #8 No Assumptions - Verify completion before marking tasks done
  • #9 Search Before Create - Context query finds prior decisions
  • #10 Complete Execution - Full 4-phase workflow from start to capture

Full Principles: CODITECT-STANDARD-AUTOMATION.md


Skill Version: 1.0.0 Created: 2026-01-01 Author: CODITECT Core Team