Skip to content

Scribe Agent ​

Reads meeting notes that Gemini saves to Google Drive after a Google Meet call, maps discussion topics to the issue backlog, and adds comments to relevant issues or creates new ones.

Setup ​

If you want to give autonomous agents access to your meeting notes, you immediately face a trust problem: how do you prevent the agent from reading notes it shouldn't have access to and then happily exposing that information in public issues?

The answer is a dedicated GCP service account. You create it in Google Cloud, and by default it has access to zero Drive files. You then invite the service account's email address to the Google Calendar events you want it to scribe. (In the calendar event settings you also need to enable Gemini notes and grant read access to attendees outside your organization.) In our experience, this calendar invite is how the resulting notes document becomes visible to the service account's Drive access — but the exact behavior may depend on your Workspace edition and admin policies (domain-wide delegation settings, external-guest sharing restrictions, etc.). Consult your Workspace admin if the service account cannot see expected notes.

At runtime, the pre-script queries the Drive API using a keyword search (SCRIBE_SEARCH_QUERY) over a rolling time window (SCRIBE_LOOKBACK_HOURS, default 3 hours) across everything the service account can see — including Shared Drives if the account has been added to any. This means the service account can read notes from any meeting it has been invited to, not just a single event. To keep the blast radius small, use a distinctive search query and avoid adding the service account to unrelated Shared Drives.

Scribe wakes up on a schedule, uses the service account credentials to search Drive for matching notes, and processes them: it files new issues on your repo or comments on existing ones, noting that the team discussed the topic in their meeting. This is an important bridge between the team's life of human interaction and the fullsend agentic system — the filed and commented-on issues serve as fodder for the triage agent, coding agent, and others.

How it helps ​

  • Meeting decisions and action items reach the issue backlog without manual copy-paste.
  • Topics are matched to existing issues by title and body content, not just keywords.
  • Public-safety and PII gates prevent confidential meeting content from reaching the issue tracker.
  • Idempotency checks avoid duplicate comments when the same notes URL was already posted.

Triggers ​

The scribe agent runs on a schedule or via manual trigger.

Commands ​

The scribe agent does not accept slash commands.

Control labels ​

Scribe does not consume or apply labels that gate agent behavior. It does apply a meeting-notes label (or agent-specified labels) to issues it creates, for categorization only.

Configuration ​

Register the agent in your .fullsend config (ADR 0058):

bash
fullsend agent add \
  https://github.com/fullsend-ai/agents/blob/main/harness/scribe.yaml \
  --name scribe \
  --fullsend-dir .

Variables ​

Per ADR 0049, scribe configuration uses the SCRIBE_ prefix.

VariableRequiredDescription
SCRIBE_REPOyesTarget repository (owner/name on GitHub, group/project on GitLab)
SCRIBE_SEARCH_QUERYyesDrive search term for meeting note doc names
SCRIBE_LOOKBACK_HOURSnoHow far back to search Drive (default: 3)
SCRIBE_DRY_RUNyestrue to preview; false for live writes
SCRIBE_MIN_CONFIDENCEnoMinimum confidence threshold (default: 0.6)
SCRIBE_MODEnoall, comments_only, or new_issues_only
FULLSEND_FORGEyesgithub or gitlab
GH_TOKENyes (GitHub)GitHub token with issues read/write
CONTENTS_TOKENno (GitHub)Optional separate token for repo content reads; falls back to GH_TOKEN
GITLAB_TOKENyes (GitLab)GitLab personal/project access token
GOOGLE_APPLICATION_CREDENTIALSyesGCP service account key for Drive read
SCRIBE_DRIVE_CREDENTIALSnoOverride path to a Drive-scoped SA key (defaults to GOOGLE_APPLICATION_CREDENTIALS)
SCRIBE_SLACK_WEBHOOK_URLnoOptional Slack notification after run

Modes ​

ModeEffect
allPost comments on existing issues and create new issues
comments_onlySkip new issue creation
new_issues_onlySkip comments on existing issues

GitLab host validation ​

gitlab-scribe-ops.lib.sh uses CI_SERVER_HOST, a GitLab CI predefined variable set automatically by the runner, as the API host. Validation fails closed when CI_SERVER_HOST is not set.

How the agent works ​

A pre-script on the host fetches open issues, recently closed issues, open PRs, and a docs index for context, then queries Google Drive for recent meeting notes. Notes are structurally scrubbed (transcript sections removed), PII patterns redacted, and packaged into the sandbox workspace.

The sandboxed agent reads the cleaned notes and repo context, extracts actionable topics, and writes validated JSON mapping topics to existing issues or new issue proposals. The agent cannot reach the forge or Drive directly — it only produces structured output.

The post-script deduplicates topics, applies confidence and public-safety gates, checks for sensitive content, and writes approved comments and issues via forge-specific API calls (gh on GitHub, curl on GitLab). Dry-run mode previews all actions without mutating the repository.

Security model ​

  • Pre-script PII scrubbing runs on the host before the agent sees notes. Bracketed Gemini attendee names ([John Smith]) and bullet attribution lines (- Jane Doe: action) are anonymized; transcript sections are dropped. Other unbracketed names in prose rely on the agent's public_safe gate as defense-in-depth.
  • Sandbox network policy allows Vertex AI only — curl is excluded to prevent exfiltration of the mapped GCP service account key.
  • Post-script gates reject topics below confidence threshold, with sensitive patterns, suspicious Unicode, or public_safe: false.
  • Dry-run gate — the post-script refuses to run unless SCRIBE_DRY_RUN is explicitly set.

Output ​

The agent produces JSON validated against schemas/scribe-result.schema.json:

  • topics[] — discussion topics mapped to existing issues (comment body in summary)
  • new_issues[] — proposals for issues not yet in the backlog
  • stats — counts for observability

Custom network policy ​

If this agent needs to reach hosts beyond the defaults, see the custom network policy guide.

Runtime support ​

Supported runtimes: claude (stable default), pi (experimental). No single-context fallback — full meeting-notes processing runs on both runtimes.

Effort: high (explicit in the harness; override per run with fullsend run --effort or FULLSEND_EFFORT, values low–max).

Source ​

harness/scribe.yaml