feat(memory-ingest): honor the per-remote deny/read-only trust policy (#2392)

Transcript ingest now respects the same trust store as code import — the gate
existed only in gstack-gbrain-sync's runCodeImport, so memory-ingest happily
ingested transcripts from deny-listed repos. preparePages filters prepared
transcript pages through ONE batch policy lookup (new 'get --batch' verb on
bin/gstack-gbrain-repo-policy — the script owns URL normalization; the client
adds repoPolicyTierBatch, one spawn for all distinct remotes, so large corpora
never pay a 10s-timeout subprocess per remote).

Outcomes match code-import semantics: read-only → clean skip
(skipped_policy_readonly), deny → counted refusal (skipped_policy_deny),
corrupted/unreadable store → HARD ERROR before any write (state, staging,
egress receipt, and import all untouched) with the recovery command named —
policy corruption must never read as successful ingestion. Artifacts are
never policy-filtered (their git_remote is a project slug, not a remote).

Fixes #2392.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This commit is contained in:
Garry Tan
2026-08-17 10:39:24 -07:00
co-authored by Claude Fable 5
parent 40e4a53f74
commit 2494276742
6 changed files with 632 additions and 2 deletions
+50 -1
View File
@@ -7,6 +7,13 @@
# if no URL is passed. Exits 0 with one of: read-write, read-only,
# deny, unset.
#
# gstack-gbrain-repo-policy get --batch
# Read remote URLs from stdin (one per line); print one tier per line
# in input order: read-write, read-only, deny, or none (no entry / no
# store). A corrupt store is a hard error (exit 2), NEVER quarantined:
# batch callers are unattended ingest gates that must fail closed
# rather than bypass a set policy.
#
# gstack-gbrain-repo-policy set <remote-url> <read-write|read-only|deny>
# Persist a tier for the given remote. Exits 0 on success.
#
@@ -161,8 +168,50 @@ ensure_file() {
fi
}
# get --batch — bulk lookup for ingest gates. One URL per stdin line, one
# tier per stdout line, input order preserved. Reuses normalize() (the same
# code path single `get` uses) per line. Prints `none` where single `get`
# prints `unset` — batch consumers (lib/gbrain-repo-policy-client.ts) speak
# the RepoPolicyTierValue vocabulary directly.
#
# Corruption polarity differs from single `get` ON PURPOSE: interactive
# `get` quarantines a corrupt store and starts fresh because /setup-gbrain
# re-asks the user; a batch caller is an unattended ingest gate with nobody
# to re-ask, so silently quarantining would BYPASS a set deny policy. Batch
# fails hard (exit 2) instead and names the recovery path.
cmd_get_batch() {
require_jq
if [ ! -f "$POLICY_FILE" ]; then
# No store = no policy was ever set. Every URL is `none`; don't create
# the file just for a read (matches cmd_list).
while IFS= read -r url || [ -n "$url" ]; do
printf 'none\n'
done
return 0
fi
if ! jq empty "$POLICY_FILE" 2>/dev/null; then
die "policy store $POLICY_FILE is corrupt (invalid JSON) — refusing batch read. Inspect with: gstack-gbrain-repo-policy list; re-run /setup-gbrain to rebuild the store."
fi
# Valid JSON from here, so ensure_file only performs the legacy
# allow → read-write migration (never the quarantine branch).
ensure_file
local url key
while IFS= read -r url || [ -n "$url" ]; do
key=$(normalize "$url")
if [ -z "$key" ]; then
printf 'none\n'
continue
fi
jq -r --arg key "$key" '.[$key] // "none"' "$POLICY_FILE"
done
}
cmd_get() {
local url="${1:-}"
if [ "$url" = "--batch" ]; then
cmd_get_batch
return 0
fi
if [ -z "$url" ]; then
url=$(git remote get-url origin 2>/dev/null || true)
if [ -z "$url" ]; then
@@ -221,7 +270,7 @@ case "${1:-}" in
set) shift; cmd_set "$@" ;;
list) shift; cmd_list "$@" ;;
normalize) shift; cmd_normalize "$@" ;;
--help|-h|help) sed -n '2,47p' "$0" | sed 's/^# \{0,1\}//' ;;
--help|-h|help) sed -n '2,54p' "$0" | sed 's/^# \{0,1\}//' ;;
"") die "usage: gstack-gbrain-repo-policy {get|set|list|normalize|--help}" ;;
*) die "unknown subcommand: $1" ;;
esac