Thread Executor

Write agent outputs to thread stage files using the numbered convention.

Responsibility

Accept agent outputs and write them to the correct numbered stage files in a thread directory. Update frontmatter to track completion.

Thread File Convention

threads/{domain}/{thread-name}/
├── 1-input.md          # Goal context (written by sys-activating-goals, NEVER by this skill)
├── 2-hypothesis.md     # Approach, assumptions, expected outcomes
├── 3-implication.md    # If hypothesis true, what follows
├── 4-decision.md       # Chosen course of action
├── 5-actions.md        # Execution steps and status
└── 6-learning.md       # Outcomes, insights, improvements

Agent-to-Stage Mapping

Type	Stages Written	Agents
planning	2, 3, 4	mkt-strategist, sls-strategist
execution	5, 6	mkt-campaign-manager, sls-outbound-manager, cst-advocacy-manager
spec	2, 3, 4, 5	prd-engineer, prd-growth-engineer
full-cycle	2, 3, 4, 5, 6	ops-manager
lifecycle	2, 4, 5, 6	cst-success-manager, cst-expansion-manager, cst-retention-manager
content	2, 5	mkt-content-manager, sls-enablement-manager
inbound	2, 3, 5	mkt-inbound-manager
partnership	2, 4, 5, 6	sls-partner-manager

Input

The calling agent provides:

thread_path: threads/{domain}/{thread-name}
agent_name: {agent that produced the output}
agent_type: planning | execution | spec | full-cycle | lifecycle | content | inbound | partnership
outputs:
  hypothesis: |    # Stage 2 content (if in agent's stages)
    ...
  implication: |   # Stage 3 content
    ...
  decision: |      # Stage 4 content
    ...
  actions: |       # Stage 5 content
    ...
  learning: |      # Stage 6 content
    ...

Process

Step 1: Validate Thread

1. Confirm thread_path exists
2. Confirm 1-input.md exists (thread was properly activated)
3. Read 1-input.md frontmatter for thread_id and goal_id

Step 2: Determine Stages

1. Look up agent_type in mapping table
2. Get list of stage numbers to write
3. Confirm agent provided content for each stage

Step 3: Write Stage Files

For each stage in the agent's mapping:

1. Create stage file with frontmatter:
   - status: completed
   - completed_by: {agent_name}
   - completed_at: {YYYY-MM-DD}
2. Write agent output as file body
3. Save file

Step 4: Verify

1. Confirm all mapped stages have status: completed
2. Return summary of stages written

Frontmatter

Stage files are created by this skill (not pre-existing). Each file gets:

---
status: completed
completed_by: mkt-strategist
completed_at: 2026-02-01
---

For 4-decision.md specifically:

---
status: completed
decided_by: mkt-strategist
decided_at: 2026-02-01
---

For 5-actions.md specifically:

---
status: completed
started_at: 2026-02-01
completed_at: 2026-02-01
---

Stage Content Guidelines

2-hypothesis.md

Agent fills:

• Approach section with strategy/methodology
• Expected Outcome with measurable targets
• Key Assumptions table
• Risks table

3-implication.md

Agent fills:

• If Hypothesis Succeeds with quantified impact
• If Hypothesis Fails with fallback plan
• Decision Criteria table with thresholds

4-decision.md

Agent fills:

• Chosen Action with explicit commitment
• Rationale linking back to hypothesis
• Trade-offs table
• Approval gate (if impact >= 0.8)

5-actions.md

Agent fills:

• Execution Plan table with steps, owners, due dates
• Progress Log (updated during execution)
• Blockers (if any)

6-learning.md

Agent fills:

• Outcome (expected vs actual)
• What Worked list
• What Didn't Work list
• Key Insights
• Process Improvements
• Recommendations

Output

execution_summary:
  thread_path: threads/{domain}/{thread-name}
  agent: {agent_name}
  type: {agent_type}
  stages_written: [2, 3, 4]  # example for planning type
  completed_at: 2026-02-01
  status: all_stages_written | partial

Integration

Upstream

• All thread-routable agents: Produce outputs then call this skill
• sys-activating-goals: Creates thread with 1-input.md (prerequisite)

Downstream

• sys-tracking-goals: Reads completed stages to assess progress
• 6-learning.md: Feeds back into goal tracking and loop detection

Constraints

This skill NEVER:

• Writes to 1-input.md (owned by sys-activating-goals)
• Creates thread directories (owned by sys-activating-goals)
• Executes domain work (agents do)
• Skips stages in the agent's mapping

This skill ALWAYS:

• Validates thread exists before writing
• Updates frontmatter status on every stage file
• Records which agent completed each stage
• Returns execution summary