architect: gatekeeper agent — external signal verification #31

Merged

disinto-admin merged 1 commit from architect/gatekeeper-agent into main

2026-04-15 17:39:03 +00:00

architect-bot commented

2026-04-15 03:03:17 +00:00

Collaborator

What this enables

The factory gains a trust boundary between external mirrors (Codeberg, GitHub) and the internal Forgejo issue tracker. Today, external bug reports and feature requests must be manually copied into internal Forgejo. The gatekeeper closes this gap: poll mirrors, verify claims against internal ground truth, create sanitized internal issues.

Supports Growth goals: attract developers, contributors, lower the barrier to entry.

Complexity

~5 new files, ~3 modified — 70% gluecode, 30% greenfield
Follows existing agent-run.sh + formula pattern exactly
No new containers, runs in existing agents container
Estimated 5-7 sub-issues

Risks

Prompt injection — mitigated by evidence-based rewriting (never verbatim copy)
AD-006 tension — gatekeeper needs READ access to external APIs (vault-only tokens)
False positives/negatives — mitigated by downstream reproduce/triage pipeline

Cost

1 new agent (6h interval), 1 new Forgejo bot, 1 new formula
No new services or containers

Recommendation

Worth it. Completes the mirror loop (outbound push exists, inbound feedback missing). Mostly gluecode. Main decision point: AD-006 token access pattern.

Reply ACCEPT to proceed with design questions, or REJECT: <reason> to decline.

## What this enables The factory gains a trust boundary between external mirrors (Codeberg, GitHub) and the internal Forgejo issue tracker. Today, external bug reports and feature requests must be manually copied into internal Forgejo. The gatekeeper closes this gap: poll mirrors, verify claims against internal ground truth, create sanitized internal issues. Supports Growth goals: attract developers, contributors, lower the barrier to entry. ## Complexity - ~5 new files, ~3 modified — 70% gluecode, 30% greenfield - Follows existing agent-run.sh + formula pattern exactly - No new containers, runs in existing agents container - Estimated 5-7 sub-issues ## Risks 1. **Prompt injection** — mitigated by evidence-based rewriting (never verbatim copy) 2. **AD-006 tension** — gatekeeper needs READ access to external APIs (vault-only tokens) 3. **False positives/negatives** — mitigated by downstream reproduce/triage pipeline ## Cost - 1 new agent (6h interval), 1 new Forgejo bot, 1 new formula - No new services or containers ## Recommendation **Worth it.** Completes the mirror loop (outbound push exists, inbound feedback missing). Mostly gluecode. Main decision point: AD-006 token access pattern. --- Reply `ACCEPT` to proceed with design questions, or `REJECT: <reason>` to decline.

architect-bot added 1 commit 2026-04-15 03:03:17 +00:00