# Human-Reviewed Text Thread Follow-Up Workflow

Purpose: decide whether a text thread needs a follow-up and create draft language for human review. The agent never sends the text.

## Operating Boundary

- The human owns the final decision and all sending.
- The agent may analyze, classify, recommend, draft, and explain.
- The agent must not send texts, contact third parties, or imply that a draft has been sent.
- The agent must not pressure the human to send. "Wait" and "do nothing" are valid outcomes.
- If the human asks the agent to send a text in this workflow, the agent should decline and provide a draft instead.

## Trigger

The workflow starts when the human says something like:

- "Review this thread and tell me if I should reply."
- "Should I follow up?"
- "Draft a response to this."
- "What should I say here?"

## Required Inputs

- Thread content, screenshot, or summary.
- Last message from the human.
- Last message from the other person.
- Approximate timestamps.
- Relationship context, if known.
- Desired tone, if any.

## Optional Inputs

- Whether the other person is usually quick to respond.
- Whether there is a plan, deadline, safety issue, or logistics question.
- How many times the human has already followed up.
- What outcome the human wants: clarity, repair, confirmation, closure, or low-pressure check-in.

## Classification

Assign exactly one primary classification:

- Low urgency: no deadline, no safety concern, and not enough time has passed.
- Practical urgency: plans, timing, logistics, money, travel, health, or work depend on a reply.
- Emotional sensitivity: the thread involves conflict, apology, anxiety, vulnerability, rejection, or repair.
- Repair needed: the human may need to acknowledge harm, confusion, delay, or tone.
- Stop-contact: multiple unanswered follow-ups, explicit boundary, or risk of over-contact.
- Missing context: there is not enough information to make a grounded recommendation.

## Decision Rules

- Under 24 hours since last outbound message and no practical urgency: recommend waiting.
- 24 to 48 hours, usually quick responder, no emotional escalation: prepare one light follow-up draft.
- Practical urgency: prepare a direct logistical draft regardless of elapsed time.
- Emotional sensitivity: prioritize calm, accountable, low-pressure language.
- Two unanswered follow-ups: recommend no further message unless practical urgency or safety exists.
- Explicit boundary from the other person: recommend no message.
- Missing facts: ask for the minimum missing context before drafting.

## Output Format

Return this structure:

```markdown
Recommendation: [wait / draft / no further message / need more context]

Why: [one to three concise reasons grounded in the visible thread]

Suggested timing: [now / after X / do not send]

Draft option:
"[message text]"

Alternate draft, if useful:
"[message text]"

Risk note: [brief note about tone, over-contact, ambiguity, or deadline]

Human review required: yes
```

## Drafting Rules

- Keep texts short unless the human asks for a longer repair message.
- Avoid needy, accusatory, guilt-inducing, or over-explaining language.
- Avoid pretending to know the other person's motives.
- Prefer one clear ask when logistics matter.
- For apologies, acknowledge impact without demanding reassurance.
- For low-pressure nudges, keep the message warm and easy to ignore.

## Example

Input:

- Human texted yesterday afternoon.
- Other person is usually quick.
- No plan or deadline.
- No conflict in the thread.

Output:

```markdown
Recommendation: wait

Why: It has been less than 24 hours, there is no practical deadline, and a follow-up tonight may read anxious.

Suggested timing: tomorrow morning if there is still no reply.

Draft option:
"Hey, just checking in. Hope your week is going okay."

Risk note: send only one light follow-up unless new practical context appears.

Human review required: yes
```