Archestra Dependency Override Sweep
CVE fixes for transitive or pinned deps — the ones Dependabot can't auto-fix — live as
overrides in platform/pnpm-workspace.yaml. Two kinds of cruft collect there, and this
skill removes them:
- Matured temporary pins. When a fix is newer than the repo's 7-day
minimumReleaseAge (10080 minutes) it's pinned exact and the package is added to
minimumReleaseAgeExclude so pnpm installs it anyway. Those are temporary — unwind
them once the fix has cleared the window.
- Redundant overrides. As the dependency graph catches up, an override stops doing
anything: the package already resolves to a compliant version without it.
Out of scope: authoring new CVE fixes (adding overrides for freshly-flagged
advisories). This skill only sweeps overrides that already exist.
Work from platform/. Make one override change at a time — one matured pin unwound
(Mode A) or one redundant override removed (Mode B), never several and never one of each.
Smallest blast radius, trivially revertible, easy to bisect.
Mode A — unwind one matured temporary pin
- Find the temporary entries: the
TEMPORARY: comment blocks and the
minimumReleaseAgeExclude list (ignore non-CVE excludes like next / @next/* —
they're kept for other reasons).
- Pick one whose pinned fix version has now been published ≥7 days ago — check
npm view <pkg> time --json rather than trusting the comment's date. If unsure,
proceed anyway: pnpm is the backstop — re-resolving rejects a still-immature version
with ERR_PNPM_NO_MATURE_MATCHING_VERSION, which simply means leave it quarantined.
- For that one package: drop it (and any
@scope/* siblings) from
minimumReleaseAgeExclude along with its TEMPORARY: comment, and relax its exact
override pin to a >= floor at the fix version — or drop the override entirely if the
graph resolves to a non-vulnerable version without it. Leave pins held for a non-CVE
reason alone.
- Verify (below).
Mode B — drop one redundant override
Pick one override to test. Remove its line from overrides: (and any comment that
documents only that line).
Re-resolve: corepack pnpm install --lockfile-only --ignore-scripts.
Judge from git diff platform/pnpm-lock.yaml. Redundant ⇒ the diff is empty, or at
most drops the override's own line from the top overrides: block / flips a specifier:
reflection. What proves the override was load-bearing is any edit under the resolution
sections — importers (dependency refs), packages (a name@version: key, e.g.
+ picomatch@2.3.2:), or snapshots (dependency edges); revert if any of those moved.
Read the whole diff — don't grep for version: alone, since a shift usually shows up as a
new/removed package key, not a version: line.
pnpm (v11) often leaves the now-orphaned override line in the lockfile's overrides:
block, so an empty lockfile diff is the normal redundant result, not a sign nothing ran.
That stale line is inert metadata — not a re-applied override — and the --frozen-lockfile
check below confirms it. Don't force a full reinstall or pnpm dedupe to purge it; that
adds unrelated graph churn for no resolution benefit.
Verify (below).
Watch for false positives: a nested override (e.g. mammoth>@xmldom/xmldom) can look
redundant only because a sibling top-level floor (@xmldom/xmldom) is what actually holds
it — removing it adds a package key / shifts a version, which the diff exposes. Treat exact
pins that hold a version down conservatively; they may be pinning out a regression.
Verify (both modes)
Notes
- After a sweep the lockfile stays at the resolved version regardless of removing the
exclusion; the exclusion only governs install-time maturity enforcement.
- Overrides come in several shapes — plain (
lodash: '>=4.18.0'), exact pins
(vite: 7.3.5), major-scoped selectors (ws@>=8, picomatch@<4), and nested
(mammoth>@xmldom/xmldom). The lockfile-diff check in Mode B is what actually proves a
removal is safe, whatever the shape.
minimumReleaseAge (7 days) is a supply-chain defense; only bypass it via the exclude
list for a known security fix, and undo it promptly — that's Mode A.
1---2name: archestra-dev-override-sweep3description: Use when asked to sweep, clean up, or revisit pnpm overrides and minimumReleaseAge exclusions in platform/pnpm-workspace.yaml — unwinding a matured temporary CVE pin once its fix has cleared the 7-day window, or removing an override the dependency graph has made redundant. This skill only sweeps existing pins; it does not author new CVE fixes.4metadata:5 source: https://skillmd.com/skills/archestra-ai/archestra-dev-override-sweep6---78# Archestra Dependency Override Sweep910CVE fixes for transitive or pinned deps — the ones Dependabot can't auto-fix — live as11`overrides` in `platform/pnpm-workspace.yaml`. Two kinds of cruft collect there, and this12skill removes them:1314- **Matured temporary pins.** When a fix is newer than the repo's 7-day15 `minimumReleaseAge` (`10080` minutes) it's pinned *exact* and the package is added to16 `minimumReleaseAgeExclude` so pnpm installs it anyway. Those are temporary — unwind17 them once the fix has cleared the window.18- **Redundant overrides.** As the dependency graph catches up, an override stops doing19 anything: the package already resolves to a compliant version without it.2021**Out of scope:** authoring *new* CVE fixes (adding overrides for freshly-flagged22advisories). This skill only sweeps overrides that already exist.2324Work from `platform/`. **Make one override change at a time** — one matured pin unwound25(Mode A) *or* one redundant override removed (Mode B), never several and never one of each.26Smallest blast radius, trivially revertible, easy to bisect.2728## Mode A — unwind one matured temporary pin29301. Find the temporary entries: the `TEMPORARY:` comment blocks and the31 `minimumReleaseAgeExclude` list (ignore non-CVE excludes like `next` / `@next/*` —32 they're kept for other reasons).332. Pick one whose pinned fix version has now been published ≥7 days ago — check34 `npm view <pkg> time --json` rather than trusting the comment's date. If unsure,35 proceed anyway: pnpm is the backstop — re-resolving rejects a still-immature version36 with `ERR_PNPM_NO_MATURE_MATCHING_VERSION`, which simply means leave it quarantined.373. For that one package: drop it (and any `@scope/*` siblings) from38 `minimumReleaseAgeExclude` along with its `TEMPORARY:` comment, and relax its exact39 override pin to a `>=` floor at the fix version — or drop the override entirely if the40 graph resolves to a non-vulnerable version without it. Leave pins held for a non-CVE41 reason alone.424. Verify (below).4344## Mode B — drop one redundant override45461. Pick one override to test. Remove its line from `overrides:` (and any comment that47 documents only that line).482. Re-resolve: `corepack pnpm install --lockfile-only --ignore-scripts`.493. Judge from `git diff platform/pnpm-lock.yaml`. **Redundant ⇒ the diff is empty, or at50 most drops the override's own line from the top `overrides:` block / flips a `specifier:`51 reflection.** What proves the override was load-bearing is any edit under the resolution52 sections — `importers` (dependency refs), `packages` (a `name@version:` key, e.g.53 `+ picomatch@2.3.2:`), or `snapshots` (dependency edges); revert if any of those moved.54 Read the whole diff — don't grep for `version:` alone, since a shift usually shows up as a55 new/removed package key, not a `version:` line.5657 pnpm (v11) often leaves the now-orphaned override line in the lockfile's `overrides:`58 block, so an empty lockfile diff is the *normal* redundant result, not a sign nothing ran.59 That stale line is inert metadata — not a re-applied override — and the `--frozen-lockfile`60 check below confirms it. Don't force a full reinstall or `pnpm dedupe` to purge it; that61 adds unrelated graph churn for no resolution benefit.624. Verify (below).6364Watch for false positives: a nested override (e.g. `mammoth>@xmldom/xmldom`) can look65redundant only because a sibling top-level floor (`@xmldom/xmldom`) is what actually holds66it — removing it adds a package key / shifts a version, which the diff exposes. Treat exact67pins that hold a version *down* conservatively; they may be pinning out a regression.6869## Verify (both modes)7071- **No new CVE.** Before editing, note the high/critical advisory set from72 `corepack pnpm audit --json` (the entries with `severity` `high`/`critical`); after73 re-resolving, take it again and confirm nothing new appeared. A new advisory → revert.74- **Lockfile + types stay sound:**75 ```bash76 corepack pnpm install --frozen-lockfile --prefer-offline --ignore-scripts # must say "up to date"77 corepack pnpm install --fix-lockfile --lockfile-only --ignore-scripts # immature-deps check, must not error78 corepack pnpm --filter @backend --filter @frontend type-check79 ```80- `pnpm audit` only sees npm-package advisories, not base-image OS packages — treat it as a81 local proxy, not the last word on the image's CVEs.8283## Notes8485- After a sweep the lockfile stays at the resolved version regardless of removing the86 exclusion; the exclusion only governs *install-time* maturity enforcement.87- Overrides come in several shapes — plain (`lodash: '>=4.18.0'`), exact pins88 (`vite: 7.3.5`), major-scoped selectors (`ws@>=8`, `picomatch@<4`), and nested89 (`mammoth>@xmldom/xmldom`). The lockfile-diff check in Mode B is what actually proves a90 removal is safe, whatever the shape.91- `minimumReleaseAge` (7 days) is a supply-chain defense; only bypass it via the exclude92 list for a known security fix, and undo it promptly — that's Mode A.