An assessment of the last 100 commits found ~45% were pure "refresh planner
priorities" bookkeeping and much of the rest was thrashing on one weak provider
(AtlasCloud text-tool-call repair) and guarding docs the loop already wrote —
motion, not progress. Two prompt-policy fixes:
PLANNER (planner.md):
- Default to NOT committing. Post the assessment and close the issue; open a
PRIORITIES.md PR ONLY when the change is MATERIAL (top item changes, an item
is added/removed, or a top item's issue closed). No PRs for reorders below
the top, reword, or "keep it current" — that churn was the loop's #1 waste.
- Add a diminishing-returns guard: don't queue the Nth doc-guard or the Nth
robustness workaround for an already-tolerated class; mark exhausted areas
needs-human and rank real-headroom capability instead.
TRIAGE (triage.md):
- Cap the AtlasCloud/plan-delegate tail-chase: another instance of a class the
agent already tolerates is NOT filed as a routine patch — comment "recurred —
capped" and, if worth more, needs-human. Real regressions (lint/tests/
govulncheck on master) and genuinely new defects still get filed.
Prompt-only; reversible. Steers the loop toward outcomes over busy-work.
Claude-Session: https://claude.ai/code/session_01CmdEY7pYmV5zzwCjNJ4ykL
Co-authored-by: Claude <noreply@anthropic.com>
Adds a deterministic reachable-CVE gate: `govulncheck ./...` on every push/PR,
failing on any reachable vulnerability EXCEPT an explicit allow-list of
known-unfixable ones. Today the allow-list holds exactly the two pgx/v4 CVEs
(GO-2026-5004, GO-2026-4518) with no upstream fix (tracked in #4556), so the
gate is green now and turns red the moment a NEW vulnerability appears.
This is the deterministic layer under the `security` loop role: the role
audits with judgment, this blocks known CVEs mechanically. Also adds
`govulncheck` to the loop-triage watch list, so a newly-disclosed CVE that
reddens the gate on master auto-files a fix issue for the loop to bump the dep.
Make `govulncheck` a required status check on master to enforce it.
Verified locally: exit 3 with only the two allow-listed IDs -> gate PASS.
Claude-Session: https://claude.ai/code/session_01CmdEY7pYmV5zzwCjNJ4ykL
Co-authored-by: Claude <noreply@anthropic.com>
govulncheck reported 30 reachable vulnerabilities. Remediation:
- Pin `toolchain go1.25.12` and build CI on Go 1.25 (lint/tests workflows):
clears ~24 Go standard-library CVEs (crypto/tls, crypto/x509, net/http,
html/template, net/url, os, …) that were present under go1.24.7.
- Bump `golang.org/x/net` v0.38.0 -> v0.55.0 and `google.golang.org/grpc`
v1.71.1 -> v1.79.3 (grpc raises the module's Go directive to 1.25).
Result: govulncheck drops from 30 -> 2. The remaining two
(github.com/jackc/pgx/v4, github.com/jackc/pgproto3/v2) have no upstream fix
and require a pgx v5 migration — tracked separately; the govulncheck gate will
follow with those explicitly allow-listed until migrated.
Note: this raises go-micro's minimum Go to 1.25 (forced by the grpc security
bump). Verified: build, go vet, and the ai/agent/flow/store/registry/broker/
wrapper/cmd + grpc/net-dependent packages pass on 1.25.12.
Claude-Session: https://claude.ai/code/session_01CmdEY7pYmV5zzwCjNJ4ykL
Co-authored-by: Claude <noreply@anthropic.com>
Codex