Score Intake

If a ~~call transcription connector (e.g. Fireflies) is available:

• Ask the user: "Would you like to pull a recent call from Fireflies, or provide a file/transcript directly?"
• If pulling from Fireflies: search for recent transcripts by rep name, date range, or keyword. Present a list and let the user pick one. Use that transcript as input — skip all file-detection preprocessing.
• If providing manually: proceed to the existing Step 1 file-detection flow.

If no connector is available, proceed directly to Step 1.

Step 1: Detect Input Type and Preprocess

Before scoring, determine what the user provided and get it into text form.

Audio or Video File (.mp3, .wav, .m4a, .flac, .ogg, .wma, .aac, .mp4, .mov, .avi, .mkv, .webm)

Tell the user:

"I'll transcribe this recording first, then score the intake call."

Invoke /legal-toolkit:transcription on the file. Once transcription completes, read the resulting transcript.txt from the work directory ({parent_dir}/{filename_without_ext}_transcript_work/transcript.txt). Use that text as the call transcript and proceed to Step 2.

PDF File (.pdf)

Attempt to extract text from the PDF. If the extracted text is mostly empty (fewer than 50 words of meaningful content), the document is likely a scanned image. Tell the user:

"This looks like a scanned document. I'll OCR it first, then score the intake call."

Delegate OCR to a subagent: launch an Agent (subagent_type: "general-purpose") with prompt: "Run /legal-toolkit:extract-text on {file_path} and write the extracted text to {parent_dir}/{filename}_ocr.txt." Once the agent completes, read the OCR output and use it as the call transcript.

If the PDF has readable text, extract it and proceed to Step 2.

Plain Text (.txt file or pasted text)

Proceed directly to Step 2 with the provided text.

Step 2: Identify Rubric

Check if the user provided a custom scoring rubric (custom categories, weights, or scoring criteria). If so, use it. If not, apply the default rubric below.

Default Scoring Rubric

Score each category 1-5. Weight indicates relative importance to the overall score.

Firms can add custom categories and adjust weights. If the firm provides a custom rubric, use it instead of the defaults.

Step 3: Score the Call

1. Scorecard

For each rubric category:

• Score (1-5)
• Evidence -- quote the specific moment in the transcript that supports the score. Use exact words. Include approximate timestamp or position if available.
• Coaching note -- one actionable sentence. Must include a specific alternative phrase the rep could have used. Never write generic feedback like "needs improvement."

Scoring anchors:

• 5: Textbook execution -- the rep did everything right for this category
• 4: Strong with minor miss -- hit the key points, missed a nuance
• 3: Adequate -- covered the basics but didn't excel
• 2: Below standard -- missed important elements or handled them poorly
• 1: Failed or absent -- the category was not addressed, or the rep actively hurt the outcome

2. Overall Score

Weighted average based on rubric weights. Display as: X.X / 5.0 with a one-sentence summary.

3. Top Strength

The single best moment in the call. Quote it. Explain why it was effective and what it likely accomplished with the prospect.

4. Top Coaching Opportunity

The single biggest thing this rep should work on. Include:

• The specific moment where it went wrong (quote it)
• What the rep should have said instead (write the exact alternative phrase)
• Why the alternative works better

5. Objections List

Every objection the caller raised, how the rep handled it, and a better response if applicable.

6. Sign Probability

Based on this call: High / Medium / Low likelihood the prospect signs. One to two sentences explaining the reasoning -- what factors in the call drive this prediction.

Accuracy and QA (Required)

Anti-hallucination rules (include in ALL subagent prompts):

• Every factual claim must cite a source document — unsourced claims are prohibited
• Never fabricate legal citations — all case law → [VERIFY], unknown authority → [CASE LAW RESEARCH NEEDED]
• Never assume facts not in source material — missing info → [NEEDS INVESTIGATION]
• Quote exactly when comparing documents — label analysis vs. facts distinctly

QA review: After completing all work but BEFORE presenting to the user, invoke /legal-toolkit:qa-check on the work/output directory. Do not skip this step.

Quality Standards

• Every score requires a quote. No score without evidence from the transcript. If a category was not addressed in the call, score it 1 and note "Not addressed in the call."
• Coaching notes must be actionable. "Needs to improve objection handling" is not acceptable. "When the caller said 'I need to think about it,' the rep should have responded: 'I completely understand -- most of our clients feel the same way before we talk through the process. Can I take two minutes to explain what happens next so you have the full picture?'" -- that is acceptable.
• Do not inflate scores. A call where the rep was "fine" is a 3, not a 4. Reserve 5 for genuinely excellent execution. If all scores cluster at 3-4, the rubric may need more specificity -- note this.
• Maintain the firm's perspective. Score against what THIS firm values, not a generic sales framework. If the firm weights qualification at 5 and closing at 2, a call with great questions but no close attempt can still score well overall.

Edge Cases

• Short calls (under 2 minutes): Score what is present. Note: "Call duration was under 2 minutes. Categories not reached due to short call length are scored 1 with the note 'call ended before this stage.' Short call duration itself may indicate a rapport or qualification failure -- assess whether the rep lost the caller early."
• Bad or partial transcripts: Score what is legible. Flag sections marked [inaudible] or [unclear]. Note: "Transcript quality affected scoring for X categories. Scores for those categories should be treated as estimates."
• Caller was hostile or unreasonable: Note the caller's behavior and adjust expectations. A rep who maintains professionalism with a hostile caller deserves credit even if they cannot close. Note: "Caller was confrontational from the outset. Scoring adjusted to reflect difficulty -- rep's composure and de-escalation efforts are evaluated."
• Caller was already sold (easy call): Note this. A call where the caller says "I want to hire you" in the first 30 seconds does not test the rep's skills. Score the categories that were still relevant (compliance, qualification, fee presentation) and note that others were not tested.
• No objections raised: Score Objection Handling as N/A rather than penalizing. Note: "No objections were raised during this call. Objection handling could not be evaluated."
• Rep gave legal advice: Flag this immediately as a compliance issue regardless of how well the rest of the call went. This is a risk item, not just a scoring issue.

Connector Action: ~~chat

If a ~~chat connector (e.g. Slack) is available, offer to post the scorecard summary to a channel:

"Want me to post this scorecard to a Slack channel for coaching review?" If yes, post a formatted summary (rep name, overall score, top strength, top coaching opportunity).