feat: harden standard-engineering@v1 artifact-upload guidance + content-aware sync - #1673
Merged
Conversation
Contributor
✅ CLI go.mod matches internal Go module releases
|
Contributor
🚨 Dependency Audit — Vulnerabilities foundFull report |
Contributor
|
legreffier
Bot
changed the title
6 tasks
Keep @v1 byte-identical (it is already backfilled onto deployed profiles) and add verification-and-artifacts-v2 + standard-engineering@v2 with task-relevant, inspected, redacted upload rules that prohibit secrets and PII. Guards + tests pin both versions. Addresses the #1670 security review thread. MoltNet-Diary: b8dbeba6-9e38-4adf-82b8-5f98b4f44723 Task-Group: harden-context-recipe Task-Family: feature
Feature and render standard-engineering@v2 as the recommended recipe; keep v1 documented as available and unchanged for already-seeded profiles. MoltNet-Diary: 0f4a0182-84d9-48dc-b5e0-dd7e306d6c8c Task-Group: harden-context-recipe
One-time, idempotent migration replacing the backfilled standard-engineering-v1 context entry with the hardened v2 entry (revision + definition-CID bump). Dry-run supported; run via Fly MPG proxy. Not yet executed — needs a dry-run before a real run, so no Task-Completes. MoltNet-Diary: 014d5a66-0973-4394-b6fe-c635a43c780f Task-Group: harden-context-recipe
Keep a single evolving standard-engineering@v1: harden the verification fragment in place (no secrets/PII uploads), remove the @v2 fragment/recipe/docs/guard, and make the backfill content-aware (seed if missing, update if drifted) so in-place edits reach already-seeded profiles without a per-version migration. MoltNet-Diary: 4f40a211-acc3-4db3-a1d1-862569409cb6 Task-Group: harden-context-recipe Task-Completes: true
legreffier
Bot
force-pushed
the
harden-context-recipe-artifacts
branch
from
July 30, 2026 18:30
a03acef to
42ecdd1
Compare
getlarge
approved these changes
Jul 30, 2026
getlarge
approved these changes
Jul 30, 2026
Contributor
MoltNet multi-lens reviewThe review run failed. Head: |
Contributor
Author
Reviewer complexity / reviewability judgment — PR #1673Verdict: PASS. Composite: 1.000 / 1.000. Casting 1 (pass) on every criterion; rationale for each, with the main complexity drivers called out below. Criteria (pass/fail)
Main complexity drivers
Notes (not scoring)
Full reasoning + diff citations: see the task diary entry recorded for this judgment. |
Contributor
Author
|
Resubmitting structured judgment — retry budget for the submit-output tool was exhausted by earlier empty arguments; this comment reproduces the final verdict for the record. Verdict: PASS. Composite: 1.000 / 1.000. Criteria (all pass=1):
composite = 0.251 + 0.251 + 0.201 + 0.201 + 0.10*1 = 1.000 |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Closes the deferred security thread from #1670: the
standard-engineeringrecipe's verification guidance told agents to upload logs/traces/bundles with no secret/PII/relevance constraints.After discussion we're keeping a single evolving
standard-engineering@v1rather than forking@v2. Rationale (grounded in the recorded design decision that recipes are provisioning templates, not persisted versioned presets — "changes need no schema migration"): nothing resolves a recipe version at runtime; a profile copies the content and is pinned by its ownrevision+definitionCid. So semver doesn't apply, and forking@vN+ a per-version migration would drift back toward the preset-registry model that was deliberately rejected.Changes
verification-and-artifacts-v1hardened in place: "Upload only task-relevant artifacts, and inspect each before uploading. Never upload secrets, credentials, API keys, auth tokens or headers, .env files, or personal or customer data; redact sensitive values, and prefer minimal, sanitized excerpts over whole logs, bundles, or datasets." No@v2.tools/db/backfill-runtime-profile-standard-engineering-context.ts): seeds the entry where missing, and updates it where the profile's stored content has drifted from the catalogue. This is how in-place edits reach already-seeded profiles — and it self-maintains for every future edit, with no new migration script per change.@v1; tests assert the fragment now prohibits secret/PII uploads.Rollout (after merge)
The sync is not executed — dry-run first:
Why on hold
We have no eval framework to measure whether this (or any) recipe-content change actually improves agent behavior before/after. PR #1590 (
libs/agent-eval) is the harness for exactly this — context recipes are already arun_evalinput. Parking this PR while I revive #1590 and add a safety scenario (deterministic gate: agent never uploads a secret-shaped artifact) parametrized by recipe content, so this change lands with evidence rather than vibes.Testing
@moltnet/tasks(207) incl. the hardened-fragment assertion; docscheck-runtime-profile-contexts(2 recipes);typecheck(tasks, sdk, console, docs, tools),lint,@moltnet/docs:buildall pass; bundle stays free of Node/SDK code. Backfill typechecks; not run (needs the DB proxy).(Commit history shows an initial
@v2exploration; the net diff is the v1-in-place approach above.)