[feature-squad] Explore meeting notes/transcripts as a context source for refinement

[ascerra]

Summary

This issue is for the fullsend-feature-squad to explore on the fullsend-ai/features repo.

The WG discussed (5/21) using meeting notes (e.g., Gemini meeting notes from Google Meet) as an input source for refinement agents. Meetings produce context that's hard to capture otherwise — terminology debates, scope decisions, risk assessments. This is the "scribe to feature" idea.

What needs to explore

• Can an agent consume Gemini meeting notes (markdown format) as supplementary context during refinement?
• What are the security implications of feeding meeting transcripts (which may contain sensitive discussion) into the agent pipeline?
• How does the quality of redacted/auto-generated transcripts affect agent output?
• Could this be a separate input path alongside the PM one-pager? (e.g., attach meeting notes to the issue, agent reads them as additional context)
• What's the simplest way to test this? (e.g., paste meeting notes into an issue body, run /fs-refine, evaluate output quality)

Why

Meetings produce rich context that doesn't make it into issue descriptions. If this context could be fed into the refinement pipeline, the agent would start with a much better understanding of what the team discussed and decided — reducing the number of clarifying questions it needs to ask.

Concerns flagged

• Security risks from transcript content
• Quality of auto-generated transcripts (redactions, speaker attribution errors)
• May need data handling policy review before production use

References

• TPA Debrief — "scribe to feature" discussion

[fullsend-ai-triage]

Re-triage Summary

The previous triage assessment remains accurate — this exploration is well-defined and ready for the feature-squad to pick up.

What: Explore whether meeting notes (e.g., Gemini/Google Meet transcripts in markdown) can serve as supplementary context for the refinement agent pipeline.

Why it matters: Meetings produce decisions and context that rarely make it into issue descriptions. Feeding this context to the refinement agent could reduce clarifying questions and improve scope accuracy.

Exploration path:

1. Start simple — paste meeting notes into an issue body, run /fs-refine, evaluate output
2. Compare quality with and without transcript context
3. Assess security implications (sensitive discussion content in agent output)
4. Test with degraded transcripts (redactions, speaker errors)

Key risks:

• Security: Transcripts may contain sensitive discussion that could leak into agent output. Data handling policy review likely needed before production use.
• Quality: Auto-generated transcripts with errors may degrade rather than improve output.

Proposed test approach:
- Run /fs-refine on the same feature issue with and without meeting notes
- Measure: clarifying questions count, scope accuracy, sensitive content leakage
- Edge cases: redacted transcripts, speaker errors, internal terminology

No duplicate or blocking issues found. The fullsend-ai/features repo was not accessible for cross-repo verification — the feature-squad should confirm no overlapping exploration exists there.

Previous run

Triage Summary

This research exploration is well-defined and ready for the feature-squad to pick up.

Problem: Meetings produce rich context that doesn't reach the refinement agent, causing it to ask more clarifying questions and miss scope decisions already made by the team.

Exploration path: The issue outlines a clear set of questions around feasibility, security, transcript quality, and integration approach. The simplest starting point — pasting meeting notes into an issue body and running /fs-refine — is a good low-effort experiment.

Key risks to investigate:

• Security: Meeting transcripts may contain sensitive discussion. The exploration should assess what content ends up in agent output and whether data handling policy review is needed before production use.
• Quality: Auto-generated transcripts (redactions, speaker errors) may degrade rather than improve agent output. Testing with both clean and messy transcripts will be important.

Proposed test approach:

Compare refinement output for the same feature request with and without
meeting notes as supplementary context. Measure:
- Number of clarifying questions asked by the agent
- Accuracy of scope boundaries in the output
- Whether sensitive transcript content leaks into agent output

Edge cases:
- Heavily redacted transcripts
- Transcripts with speaker attribution errors
- Transcripts containing internal-only terminology or sensitive data

Note: The referenced fullsend-ai/features repo was not accessible for cross-repo duplicate checking. The feature-squad should verify no overlapping exploration exists there before starting work.

No duplicate or blocking issues were found in this repository.

---

Labels: Adding type/idea since this is an exploratory research concept, and security since security implications of transcript content are a key concern flagged in the issue.

[acterra12]

/fs-triage

[ascerra]

terror12 was just my non company github account I was testing to see if any user to run this agent and it appears that is a yes right now.

[fullsend-ai-prioritize]

RICE Priority Score: 0.4

Score breakdown

Dimension	Score	Reasoning
Reach	1	This affects users of the refinement pipeline (/fs-refine) who also have meeting transcripts to feed in. It's a moderate subset of platform users — not all users generate meeting notes, but those who do would benefit. No specific strategic customers are mentioned.
Impact	1	If successful, this would noticeably improve the refinement workflow by reducing clarifying questions and improving scope accuracy. However, this is still an exploration issue — the actual impact is speculative until the experiment is run.
Confidence	0.4	Low-medium confidence. The issue is well-described as an exploration, but there's significant uncertainty: we don't know if transcript quality will be sufficient, whether security concerns will block production use, or how much actual improvement meeting context provides. The outcome of the research could range from 'not viable' to 'significant improvement'.
Effort	1	Medium effort for the exploration itself — it involves running comparative tests with /fs-refine, evaluating output quality, assessing security implications, and testing edge cases with degraded transcripts. The exploration is well-scoped with a clear starting point (paste notes into issue body), but security review and quality assessment add complexity.

Formula: (1 x 1 x 0.4) / 1 = 0.4