interview-scorecard

name	interview-scorecard
description	Draft concise interview scorecards from pasted long-form interview notes, nearby Google Calendar interview context, and memory/miclog/ transcript excerpts. Use when writing hiring feedback, interview scorecards, spikes/troughs, or candidate evaluation summaries.

Interview Scorecard

Draft a concise interview scorecard from pasted interview notes, nearby calendar context, and timestamped transcripts in memory/miclog/ generated by miclog.

Required Inputs

The user should paste long-form notes from their interview. If they do not provide notes, ask for them before drafting.

If the user does not provide all three of these exact inputs, ask for the missing items before drafting:

• My high level summary is
• I want to flag that
• I recommend that we

Do not draft the scorecard until the interview notes and all three user-provided framing inputs are available.

Context Gathering

1. Look at the user's calendar for the event that is just ending, in the closing minutes, or most recently ended.
2. Prefer an interview-like event if multiple recent events match. Interview-like signals include titles or descriptions containing interview, candidate, hiring, screen, onsite, phone screen, or debrief.
3. If using the local Sidekick client, list nearby events around now:

python3 -m sidekick.clients.gcalendar list <time-min> <time-max> 20

Use a narrow window such as the last 2 hours through the next 15 minutes. Fetch full event details when useful:

python3 -m sidekick.clients.gcalendar get <event-id>

4. Use memory/miclog/AGENTS.md to find timestamped transcript files, then select lines overlapping the event window, with a small buffer before the event start and after the event end. Miclog timestamps are local and look like [YYYY-MM-DD HH:MM:SS].
5. Use pasted notes as the primary source. Use calendar and miclog context only to recover meeting context and useful quoted phrases.
6. If calendar or miclog context is unavailable, continue from the pasted notes and do not invent missing evidence.
7. Track the selected calendar item title and the number of miclog lines used so they can be reported with the result.

Drafting Rules

• Return only the context-used line and scorecard text. Do not write files, update calendar, send email, post to Slack, or create external drafts unless the user explicitly asks.
• Keep the scorecard body at or under 200 words. Do not count the context-used line toward this limit.
• Use past tense.
• Refer to the candidate only as C.
• Use an informal, straightforward, highly pragmatic style.
• Keep every sentence concise.
• Include short quoted phrases from the interview notes or miclog.
• Do not use the candidate's name, interviewer names, or unnecessary calendar metadata in the scorecard body.
• Do not include raw transcript dumps.
• Do not include caveats about missing calendar or miclog context in the final scorecard.

Output Format

Use exactly this structure:

Context used: Calendar item: [matched calendar item title, or "not found"]; Miclog lines: [number]

[One high-level summary paragraph.]

Spikes
- [Spike with evidence and a short quote.]
- [Optional additional spike.]

Troughs
- [Trough with evidence and a short quote.]
- [Optional additional trough.]

Use 1-3 bullets in each section. If the notes are thin, prefer one strong bullet over padded feedback.