Protocol Injection Flow

How protocols are loaded, combined, and injected into subagent prompts

Protocol Injection Flow

Version: 1.0.0 | Status: ACTIVE

This document describes how protocols are loaded, combined, and injected into subagent prompts.

Overview

Task Request
    │
    ▼
┌─────────────────────────────────────┐
│ 1. Skill Selection                  │
│    skill_auto_dispatch(task_id)     │
│    → Returns: ct-research-agent     │
└─────────────────────────────────────┘
    │
    ▼
┌─────────────────────────────────────┐
│ 2. Template Loading                 │
│    Load skills/ct-{skill}/SKILL.md  │
└─────────────────────────────────────┘
    │
    ▼
┌─────────────────────────────────────┐
│ 3. Base Protocol Loading            │
│    Load _shared/subagent-protocol-  │
│    base.md (OUT-001 to OUT-004)     │
└─────────────────────────────────────┘
    │
    ▼
┌─────────────────────────────────────┐
│ 4. Conditional Protocol Loading     │
│    Load applicable protocols based  │
│    on task context                  │
└─────────────────────────────────────┘
    │
    ▼
┌─────────────────────────────────────┐
│ 5. Token Resolution                 │
│    ti_set_full_context()            │
│    Replace all {{TOKENS}}           │
└─────────────────────────────────────┘
    │
    ▼
┌─────────────────────────────────────┐
│ 6. Prompt Assembly                  │
│    Combine: skill + base + cond.    │
│    + resolved tokens                │
└─────────────────────────────────────┘
    │
    ▼
Combined Prompt → cleo-subagent

Step 1: Skill Selection

The dispatch algorithm selects the appropriate skill:

skill_auto_dispatch("T1234")
# Checks: category → type → keywords → labels → default
# Returns: skill name (e.g., "ct-research-agent")

See SKILL-DISPATCH-ALGORITHM.md for details.

Step 2: Template Loading

Load the skill’s SKILL.md template:

# Path: skills/ct-{skill}/SKILL.md
template=$(cat "skills/ct-research-agent/SKILL.md")

The template contains:

• Skill purpose and capabilities
• Methodology instructions
• Token placeholders (e.g., {{TASK_ID}})

Step 3: Base Protocol Loading

Always loaded for every subagent:

# Path: skills/_shared/subagent-protocol-base.md
base_protocol=$(cat "skills/_shared/subagent-protocol-base.md")

Base Protocol Rules (OUT-001 to OUT-004):

Rule	Requirement
OUT-001	MUST write output to claudedocs/agent-outputs/
OUT-002	MUST append ONE line to MANIFEST.jsonl
OUT-003	MUST return summary message only
OUT-004	MUST NOT return content in response

Step 4: Conditional Protocol Loading

Loaded based on task context:

Protocol	Trigger Condition
Task Lifecycle	Always (task system integration)
Research Linking	Task has research dependencies
Verification Gates	Task requires verification
Phase Awareness	Project has phase tracking
Dependency Context	Task has dependencies
Error Handling	Always (fallback behavior)
Session Integration	Session is active

# Example: Load verification gates if task needs verification
if task_requires_verification "$task_id"; then
    conditional+=$(cat "skills/_shared/verification-gates.md")
fi

Step 5: Token Resolution

Replace all {{TOKEN}} placeholders with values:

ti_set_full_context "$task_id"

Token Categories:

Category	Tokens	Source
Task	{{TASK_ID}}, {{TASK_TITLE}}, {{TASK_DESCRIPTION}}	Task data
Context	{{EPIC_ID}}, {{SESSION_ID}}, {{DATE}}	Session/environment
Output	{{OUTPUT_FILE}}, {{TOPIC_SLUG}}	Generated
Manifest	{{MANIFEST_SUMMARIES}}, {{PRIOR_RESEARCH}}	MANIFEST.jsonl

Resolution Process:

# 1. Load token definitions
_ti_load_tokens_from_json()

# 2. Set task-specific values
TI_TASK_ID="T1234"
TI_TASK_TITLE="Research authentication"
TI_DATE="2026-01-27"

# 3. Pattern substitution (bash parameter expansion)
output="${template//\{\{TASK_ID\}\}/$TI_TASK_ID}"

Step 6: Prompt Assembly

Combine all components:

final_prompt="$skill_template

$base_protocol

$conditional_protocols

## Task Context
- Task ID: $TI_TASK_ID
- Session: $TI_SESSION_ID
- Date: $TI_DATE
"

Output Structure

The assembled prompt is returned as JSON:

{
  "success": true,
  "result": {
    "taskId": "T1234",
    "template": "ct-research-agent",
    "topicSlug": "research-authentication",
    "outputFile": "2026-01-27_research-authentication.md",
    "spawnTimestamp": "2026-01-27T12:00:00Z",
    "instruction": "Use Task tool to spawn subagent with the following prompt:",
    "prompt": "... combined prompt content ..."
  }
}

Error Handling

Error	Cause	Resolution
Empty prompt	Token injection failed	Check required tokens set
Missing template	Skill not found	Verify skill exists in manifest
Validation error	Invalid token patterns	Check token format

Debugging

# Enable verbose token injection
export TI_DEBUG=1
cleo orchestrator spawn T1234

# Check token values
ti_list_tokens

Related

• Skill Dispatch Algorithm
• Manifest Schema
• Skill Development Guide