Screen resumes and draft replies

Why this matters

Inbound resumes pile up. The good ones look the same as the mediocre ones in the first three seconds, the standouts get missed, and the rejection emails sit unsent because writing them feels worse than writing the job description did. So the pile grows. Real candidates lose interest. The role stays open longer than it should.

The fix isn’t better screening criteria; it’s an agent that does the labor part: read every resume against your job description, score against the criteria you actually care about, surface standouts and red flags, and draft the response in your voice. Your judgment stays in the loop for the calls that matter; the labor of getting candidates to that point lives somewhere else. Today, you build that.

You’re already using the new-hire framing for the agents on your team. This is the agent that helps you hire real ones.

Quick start

1. Sign in: open Auxot in your browser and log in.

2. Open chat with the Admin Agent: click Chat in the left menu, make sure the agent picker reads “Admin Agent.”

3. Build the hiring-triage agent: paste this:

   I want a "Hiring Triage" agent for the [role] position. When I paste in a resume, it: (1) scores fit against the job description on the hard requirements I care about (1-5), (2) scores soft signals — clarity of communication, growth indicators, ownership patterns (1-5), (3) flags standouts and red flags explicitly, (4) drafts both an advance response (invite to first conversation) and a decline response (kind, brief), and (5) recommends advance, decline, or maybe-borderline. Build it.
   

4. Answer the Admin Agent’s questions: what the job description is, what hard requirements actually matter (vs nice-to-haves), what red flags to watch for, and your tone in candidate emails.

5. Test on real recent applicants: paste in 5 resumes from your last hiring round. Compare its scores and recommendations to what you actually did.

Done? The agent is in Settings → Agents and selectable from the agent picker. Optional next step: attach your job description and “what good looks like” examples as context files (Add your first context file): this is the difference between generic and useful.

The agent can do that?

Hiring screening that just regurgitates the job description isn’t useful. These three prompts turn the agent from a keyword matcher into something that surfaces signal you’d otherwise miss.

1. Have the agent score on hard criteria AND soft signals separately

Open chat with the hiring-triage agent and ask:

For every resume, give me two scores instead of one:
- Hard fit (1-5): do they meet the technical/experiential requirements in the job description?
- Soft signals (1-5): clarity of communication in the resume itself, indicators of growth/ownership/initiative (e.g., promotions, took on something not in their job description, or fixed a problem nobody asked them to), red flags (job-hopping that looks unhealthy, gaps without explanation, or vague accomplishments).
Don't collapse them. A 4-on-hard / 2-on-soft tells me something different than a 2-on-hard / 4-on-soft. Show your reasoning for each in one line.

Why it’s non-obvious: Most resume-screening tools collapse fit into a single number, which loses the most useful signal. The strongest hires are usually high on both axes; the most underrated hires are high on soft, lower on hard (you can teach skills; you can’t teach judgment). Splitting the scores lets you find the candidates worth a conversation that a single-number system would have buried.

2. Have the agent draft both responses, every time

After the first batch:

For every resume, draft both response variants — the advance email (invite to a first conversation, warm but specific to what stood out about them) and the decline email (kind, brief, doesn't waste their time with vague "we received many strong applications" boilerplate). I'll pick one. Even when you recommend a clear advance or decline, draft both — sometimes I'll override your recommendation and I want the response ready.

Why it’s non-obvious: Decline emails sit in drafts forever because writing them feels worse than writing the job description. Pre-drafting both variants means you click send instead of stalling. Candidates get closure within 48 hours instead of 6 weeks. Your reputation as a place that responds to applicants compounds: better candidates recommend you to other candidates.

3. Build a “Job Description Refinement” that checks which requirements actually predicted strong hires

Inverted-usage move. After 90 days of using the hiring-triage agent (and ideally after a few hires have been on the team for a while), open chat with the Admin Agent:

Build me a "Job Description Refinement" agent. Its job is to look at the candidates I advanced from the hiring-triage agent over the last [N months], the ones I actually hired, and how they're performing 90 days in. Cross-reference against the job description: which requirements correlated with strong hires, which were noise, which red flags actually predicted problems vs which were false alarms? Recommend specific job description edits for the next round of hiring.

Why it’s non-obvious: Most job descriptions are written once and copied between roles for years. Half their requirements don’t actually predict performance. The job-description refinement agent turns hiring into a feedback loop: your next job description is better because the last hiring round taught you something. Pairs well with Set up your Monday morning briefing; the refinement findings can become a section of your briefing every quarter.

Go deeper

The new-hire framing connects directly to this

The agents you build for your business have descriptions, context files, and things they should never do. Hiring is the parallel exercise for human teammates: a job description is a description, the team’s working norms are the context file, the “what could break on day one” framing applies to both. The hiring-triage agent exists in this overlap: you’re using an agent to help you recruit a human one. The same care you’d give an agent’s description should show up in the job description; if it doesn’t, the triage can only be as sharp as the input.

The context files that actually matter

What separates a useful hiring-triage agent from a generic one is the context files:

• The job description itself: but written tightly, with explicit must-haves vs nice-to-haves.
• “What good looks like”: descriptions of your strongest current hires in this role, what made them strong (not their resumes; the actual qualities). Anonymize if needed.
• Past hiring traps: patterns that looked good on paper and went sideways, with one-line context for why.
• Your team’s working style: async vs sync, autonomy vs structure, and what kind of people thrive vs flame out.

Add your first context file covers context file mechanics. The “what good looks like” file matters most: without it, the agent can match keywords against the job description but has no model of what actually succeeds in your environment.

Don’t auto-send decline emails without a glance

Tempting, but bad idea. Two reasons: (1) the agent will occasionally miscategorize a candidate as decline who you’d actually want to advance: a 5-second scan of the recommendation list is cheap insurance, and (2) generic decline boilerplate is exactly the thing that gives automated hiring a bad reputation. Skim the agent’s drafted decline email before it goes out; tweak the personalization. The human-in-the-loop pattern from Set up a content review pipeline fits here: the agent drafts, you approve.

Email wiring (Gmail and Outlook)

For Gmail, wire Google’s official Gmail MCP. It is free and draft-only by design: each candidate response (advance or decline) lands in your Gmail Drafts folder; you skim, tweak personalization, and click send. The Drafts folder is the structural review gate. See Wire Gmail drafts through Google’s official MCP for the full wiring.

For Outlook, route through Composio Outlook with the agent’s tool policy restricted to draft tools only (Create email draft, Create draft reply). Exclude the send tools from the policy explicitly; otherwise the agent can send autonomously. See Lock Outlook drafts via a draft-only tool policy for the full wiring.

This applies to both advance and decline emails. The “Don’t auto-send decline emails without a glance” rule above is the same rule, but structurally enforced instead of left as a discipline you have to remember. Keep send_email out of your agent tool policies walks through why the send tool should never be on a candidate-response agent’s policy.

The “maybe” pile is where the leverage is

Clear advances and clear declines are easy. The candidates who score 3-on-hard / 4-on-soft (or vice versa) are where most of the actual hiring decisions live. Have the agent pull these out as a separate “needs your judgment” pile. Read those carefully. The clear advances need 30 seconds; the clear declines need 5 seconds; the maybes get the time you saved on the other two.

Troubleshooting

• Every resume scores 4 or 5 on hard fit. Your job description’s hard requirements are too generic. Edit the job description context file to include explicit examples of what does and doesn’t meet each requirement.
• The agent rejects candidates you would have advanced. It’s pattern-matching too narrowly to the job description wording. Add the “what good looks like” context file (see “context files that actually matter”) so it has a model of success, not just a model of the job description.
• Decline emails sound like every other AI rejection. Add 3-5 examples of decline emails you’ve sent and were happy with as a context file. The agent calibrates voice from examples, not descriptions.
• The agent flags non-traditional career paths as red flags too often. That’s the bias-check moment from the Worth knowing aside. Either the agent’s instructions are too rigid about resume conventions, or your job description itself encodes that bias. Both are fixable; both need explicit attention.

Variations & edge cases

• Free tier: the agent works at all tiers. Multi-team setups (Set up multi-team isolation) might want one hiring-triage agent per team if hiring criteria differ across roles.
• Confidential / executive search: Auxot can be self-hosted on your own infrastructure (Connect a GPU worker and View and manage your License) so resume content never leaves your environment.
• Senior vs junior roles: the soft signals matter more for senior, the hard fit matters more for junior. Tell the agent in its description which role this is and adjust scoring weight accordingly.
• Internal candidates: different agent. Internal moves are about fit-with-team and growth trajectory more than resume content. Don’t put external and internal applicants through the same screener.
• High-volume hiring (50+ applicants per role): sampled review (Set up a content review pipeline) on the bottom 30% of the agent’s recommendations: auto-decline most, randomly route 10% to you for sanity check. Catches false-decline patterns without making you read 50 declines.
• When NOT to use this: roles where the resume isn’t predictive (creative roles judged by portfolio, sales roles judged by interview performance, or anything where the work sample matters more than the credentials). The agent can still help with logistics, but the screening signal lives elsewhere.

Walkthrough

Step 1: Open chat with the Admin Agent

Click Chat in the left menu. Make sure the agent picker at the top reads “Admin Agent.”

Step 2: Prepare the context files

Before building the agent, prepare:

• The job description, with explicit must-haves vs nice-to-haves marked.
• “What good looks like”: 1-2 paragraphs describing your strongest current hires in this role, what makes them strong (qualities and behaviors, not resume bullets).
• Past hiring patterns to avoid: short list of “looked good on paper, didn’t work out” patterns you’ve seen.
• Sample decline emails: 3-5 declines you’ve sent that were warm, brief, and not boilerplate.

This 30-minute prep is what makes the agent useful instead of generic.

Step 3: Build the hiring-triage agent

Paste this:

Build me a "Hiring Triage" agent for the [role] position. When I paste in a resume, it does the following:

1. Hard fit score (1-5) against the job description's must-haves, with one-line reasoning.
2. Soft signals score (1-5) — communication clarity in the resume, growth/ownership indicators, and red flags. One-line reasoning each.
3. Standouts — anything notably impressive or unusual.
4. Red flags — anything that warrants caution.
5. Recommendation: Advance / Decline / Maybe (needs human judgment).
6. Drafts of both response variants — advance email (warm, specific to what stood out) and decline email (kind, brief, never boilerplate). Always draft both, even for clear cases.

Attach the context files I'll provide (job description, "what good looks like," past patterns, and sample declines). Tone for emails: [warm/professional/etc]. Make it ready to use.

The Admin Agent will ask which files to attach and confirm specifics.

Step 4: Test on real recent applicants

Pick 5-10 resumes from your last hiring round where you remember what you decided and (ideally) how the hires worked out. Paste them one at a time into chat with the hiring-triage agent. Verify:

• Did it correctly identify the candidates you advanced?
• Did it correctly decline the ones you declined?
• For “maybe” calls: does its judgment match yours, or does it surface something you missed?
• Are the standout/red-flag callouts real signal, or generic?
• Are the drafted emails usable, or would you rewrite from scratch?

Adjust the agent’s description and context files for any miscategorizations. The “what good looks like” file is usually where the fix lives.

Step 5: Adopt for live hiring

For the next round of inbound resumes:

• Paste each into the hiring-triage agent (or wire it to your applicant tracking system via MCP server, Add an MCP server).
• Read the recommendation list: clear advances, clear declines, maybes.
• Spend the saved time on the maybes: they’re where your judgment matters most.
• Skim the drafted decline emails before they go out. Tweak personalization. Send.
• For advances, schedule the first conversation while the agent’s context-specific intro email is fresh.

Step 6: (Optional) Build the Job Description Refinement after a few hiring rounds

Use power move 3 once you’ve completed 1-2 hiring rounds and the new hires have been on the team for 90+ days. The job-description refinement agent tells you which job description requirements actually predicted performance and which were noise. Update the job description. The next round’s triage gets sharper.

What’s next

• → Add your first context file. Job description, “what good looks like,” and sample decline emails all live as context files.
• → Prep for your next sales call. Once resumes are sorted, the live call is where your narrative and objection handling earn out.
• → Add an MCP server. Wire the hiring-triage agent directly into your applicant tracking system instead of pasting resumes manually.
• → Set up a content review pipeline. For high-stakes outbound (executive declines, etc.) the human-in-the-loop pattern fits.
• → Wire Gmail drafts through Google’s official MCP. The canonical wiring when candidate responses go through Gmail.
• → Lock Outlook drafts via a draft-only tool policy. The canonical wiring when candidate responses go through Outlook.
• → Keep send_email out of your agent tool policies. The structural argument for why the hiring-triage agent should never have a send tool, especially for declines.
• → Onboard a new teammate to your agents. Once you’ve hired the human, this is how they meet your agents.

Reference

• Two scores, not one: hard fit and soft signals separate. Strongest hires are high on both; underrated hires are high on soft, lower on hard.
• Always draft both response variants: pre-drafted decline emails actually go out; draft-on-demand ones don’t.
• Context files determine quality: job description alone isn’t enough; “what good looks like” is what teaches the agent your team’s actual success patterns.
• Bias check is a human job: sample the bottom of the pile, blind-screen periodically, treat agent/human disagreement as data.
• The “maybe” pile is where leverage is: clear advances and declines are easy; the candidates needing judgment are where you should spend your saved time.
• Job Description Refinement loop turns each hiring round into improvement for the next.
• See also: Turn meeting notes into action items and decisions, Prep for your next sales call, Add your first context file, Set up a content review pipeline, Onboard a new teammate to your agents.