Gaj Respond

The user will provide one of:

• Pasted message text (LinkedIn, email, etc.)
• Screenshot description or forwarded email content
• Summary of what the recruiter said

Extract from the message: recruiter name, company, role, platform (LinkedIn/email), and any compensation details mentioned.

Step 2: Identify or create the pipeline job

Search the pipeline for the company:

npx tsx scripts/pipeline-cli.ts search '<company name>'

If a match exists, use that job ID. If multiple matches, show them and ask the user to pick.

If no match, offer to create a new pipeline entry:

npx tsx scripts/pipeline-cli.ts add '{"company_name":"<company>","job_title":"<role>","status":"pending-review","source":"recruiter"}'

Step 3: Log the inbound message

Store the recruiter's message in the correspondence table:

npx tsx scripts/pipeline-cli.ts log-message '{"job_id":<id>,"direction":"inbound","platform":"<linkedin|email>","sender_name":"<recruiter name>","message_text":"<their message>"}'

Step 4: Detective research (via sherlock)

Before assessing interest or drafting any response, run the sherlock investigation.

Check for existing findings:

npx tsx scripts/pipeline-cli.ts get-job <job-id>

If job_data.sherlock exists and investigated_at is < 24 hours old, use the existing findings. Skip to Step 4b.

If no findings exist, run a full sherlock investigation using the process defined in @prompts/sherlock-system.md. Use these inputs:

• Company name from the pipeline entry
• Recruiter name from the inbound message
• Any URL or JD text the user provided

Launch the 5 investigation dimensions as parallel subagents per the sherlock system prompt. Store findings:

npx tsx scripts/pipeline-cli.ts update <id> 'job_data' '<sherlock findings JSON>'

Present a condensed detective report to the user (sections: company profile, recruiter profile, red flags, green flags, compensation analysis). This is for the user's eyes only.

CRITICAL: The detective report is a secret weapon for the user only. Never reveal research findings in any outbound communication. Outbound messages must reference ONLY information the recruiter explicitly provided. Do not quote back salary numbers (forces re-anchoring), hint at knowing the unnamed client, or reference Glassdoor/market data. Ask open questions that make the recruiter reveal information rather than showing you already know it. The edge only works if it stays hidden.

Step 4b: Gather context

User preferences:

• Check if ~/gaj/context/about-me.md exists. If yes, read it for role targets, salary floor, tech stack, and evaluation criteria.
• If it does not exist: ask the user for their role targets, salary floor, and key tech preferences. Mention they can set these up permanently later.

Conversation history:

• Check for prior correspondence on this job:

npx tsx scripts/pipeline-cli.ts get-correspondence <job-id>

If prior messages exist, factor them into the response (don't repeat questions already answered, acknowledge previous exchanges).

Step 5: Assess interest (informed by detective research)

Apply the 5 criteria from @prompts/recruiter-response-system.md:

1. Role fit
2. Tech stack match
3. Company stage
4. Compensation
5. Mission alignment

Classify as: strong-interest, curious, warm-decline, or hard-pass.

For hard-pass: explain the assessment to the user, do not draft a response, and offer to mark the job as filtered.

Step 6: Draft the response

Read @prompts/recruiter-response-system.md for generation rules. Read @prompts/response-templates.md for the structural scaffold matching the classification. Apply @prompts/writing-rules.md to all output.

Draft a response using the appropriate mode:

• strong-interest -> Mode 1 (Strong Interest)
• curious -> Mode 2 (Curious but Cautious)
• warm-decline -> Mode 3 (Warm Decline)

Step 7: Present the draft

Show the user:

1. Interest assessment: which classification and why (list criteria met/unmet)
2. Word count (must be within mode constraints)
3. The draft message
4. Any qualifying questions recommended for a call (for strong-interest/curious)

Step 8: Offer refinement

After presenting, offer:

• "Adjust tone" (more casual, more formal, more direct)
• "Add questions" (include specific qualifying questions)
• "Remove questions" (trim to just the response)
• "Switch mode" (reassess at a different interest level)
• "Shorten" (cut further)

Iterate until the user approves.

Step 9: Store the approved response

After user approval, log the outbound draft:

npx tsx scripts/pipeline-cli.ts log-message '{"job_id":<id>,"direction":"outbound","message_text":"<approved text>","ai_draft":"<original draft>","tone":"<tone used>","classification":"<classification>"}'

Step 10: Offer status update

If the interest assessment suggests a status change, offer it:

• strong-interest -> suggest approved status
• curious -> keep at pending-review
• warm-decline -> suggest rejected status

npx tsx scripts/pipeline-cli.ts status <id> <new-status>

CLI reference

npx tsx scripts/pipeline-cli.ts search '<company>'
npx tsx scripts/pipeline-cli.ts add '<json>'
npx tsx scripts/pipeline-cli.ts log-message '<json>'
npx tsx scripts/pipeline-cli.ts get-correspondence <job-id>
npx tsx scripts/pipeline-cli.ts status <id> <new-status>

Follow-up mode

If the user says "follow up with [recruiter]" or "they haven't responded":

1. Look up prior correspondence via get-correspondence
2. Check how long since the last outbound message
3. If 5+ business days: use Mode 4 (Follow-up After Silence) from the response templates
4. If fewer than 5 days: suggest waiting