Skip to content

OpenShell v0.1.8 compatibility (tracking) - #82

Draft
ilanh wants to merge 11 commits into
mainfrom
openshell-next
Draft

ilanh wants to merge 11 commits into
mainfrom
openshell-next

Conversation

@ilanh

@ilanh ilanh commented Nov 9, 2026

Copy link
Copy Markdown
Owner

Tracking PR for #80. Draft: do not merge until OpenShell v0.1.8 is stable. Promotion steps are in docs/RELEASING.md (added in the automation PR for #79).

openshell-next = main + the OpenShell pin moved to the newest upstream tag, prereleases included. Once #79 merges, .github/workflows/openshell-next.yml will merge main into it daily, move the pin on new upstream tags and report on #80.

Now: openshell-policy-schema and openshell-policy are on v0.1.8-pre.3 (53b8eec2, compare with v0.1.7). Nothing broke: fmt, clippy -D warnings, cargo test --locked, gateway_policies, gen-docs.sh --check and gen-actions-docs.sh --check all pass locally with no code changes.

Before promoting (checklist mirrors #80)

Pins openshell-policy-schema and openshell-policy to v0.1.8-pre.3 (53b8eec2). Upstream changes: NVIDIA/OpenShell@v0.1.7...v0.1.8-pre.3

Refs #80

Signed-off-by: Ilan Harel <ilan@Ilans-MacBook-Pro.local>
@ilanh ilanh added the openshell-next Tracking the next OpenShell release on the openshell-next branch label Nov 9, 2026
ilanh added a commit that referenced this pull request Nov 10, 2026
…errors (#84)

Closes #83

The first manual run of the `openshell-next` workflow
([42636278277](https://github.com/ilanh/openshell-annotate/actions/runs/42636278277))
went green while doing nothing: the branch has no copy of the scripts,
and `| tee` hid the failures.

## Changes
- **`defaults: run: shell: bash`** in `openshell-next.yml` runs every
step with `-eo pipefail`, so the redundant `set -o pipefail` in the test
step is gone. I checked the other workflows for the same `| tee`
problem: composite actions already set `shell: bash`, `ci.yml` and
`upstream-compat.yml` don't pipe into `tee` (`upstream-compat.sh` checks
`PIPESTATUS`), and the pipes in `release.yml` sit inside `if`/`||`
checks that fail correctly.
- **The tooling always comes from main.** A new step copies
`scripts/openshell-next.sh` and `scripts/openshell-tags.sh` to
`$RUNNER_TEMP/tools` before checking out `openshell-next`, then runs
them against the checkout with
`OPENSHELL_ANNOTATE_ROOT=$GITHUB_WORKSPACE`. Both scripts now honor that
env var; `openshell-next.sh` passes it on to `openshell-tags.sh` and
calls it by absolute path. A fix to the scripts now applies on the next
run.
- **The report reflects reality.**
- It reads the pin from the checkout, falling back to the bump output
and then to the pin at checkout, so it no longer shows `unknown`.
- It adds a "Pin update" row and a ❌ in the heading when any step
failed, plus a tail of the bump log when the bump fails.
  - `tail` of a missing log no longer aborts the comment under `-e`.
  - The final step fails the run on a bump failure too, with `always()`.
- **Docs:** `docs/DEVELOPING.md` describes the main-tooling and pipefail
behaviour and `OPENSHELL_ANNOTATE_ROOT`, and CHANGELOG has a new line.
The AGENTS.md mapping row and RELEASING are unchanged because nothing
they describe changed. `gen-actions-docs.sh --check` passes; the
composite actions are untouched.

## Verified (on nemoclaw, in a temp clone)
- actionlint 1.7.7 is clean, and shellcheck 0.10.0 is clean on
`openshell-{next,tags}.sh` and `upstream-compat.sh`.
- **The scenario that failed:** I ran main's scripts from a temp dir
against a checkout of `origin/openshell-next` (`cea9e1e`, no scripts).
  - `sync origin/main` merged cleanly.
- `update` moved the pin from v0.1.8-pre.3 to the newly tagged
**v0.1.8-pre.4**.
- On pre.4, fmt, clippy `-D warnings`, `cargo test --locked`,
`gen-docs.sh --check` and `gen-actions-docs.sh --check` all pass.
- **The report step**, extracted from the YAML and run with `bash -eo
pipefail` and a stubbed `gh`:
- Bump failure: heading ❌, "Pin update ❌ failed", log tail, pin
`v0.1.8-pre.3`.
  - Forced report with no change: "already on the newest tag".
- Merge and tests failed with no logs: the comment is still posted ("(no
log captured)"), with exit 0.

After this merges, one manual run should merge main into
`openshell-next`, move it to v0.1.8-pre.4, push, dispatch CI on #82 and
post a real comment on #80.

Signed-off-by: Ilan Harel <ilanharel@gmail.com>
Signed-off-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Automated by .github/workflows/openshell-next.yml. Upstream changes: NVIDIA/OpenShell@v0.1.8-pre.3...v0.1.8-pre.5

Signed-off-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
ilanh added a commit that referenced this pull request Nov 11, 2026
…ll-next (#86)

Closes #85

## Cause
Since
[2026-06-11](https://github.blog/changelog/2026-06-11-bot-created-pull-requests-can-run-workflows-if-approved/),
GitHub creates `pull_request` runs for a PR updated with `GITHUB_TOKEN`
in an approval-required state
([docs](https://docs.github.com/en/actions/concepts/security/github_token)).
This doesn't depend on the fork / first-time-contributor setting, which
here is `first_time_contributors`.
- The `openshell-next` workflow pushes as `github-actions[bot]`, so CI,
Upstream compatibility and Release on #82's head (`2fbb6e7`) are
`action_required` with no check runs.
- The dispatched CI run's successful `check` / `audit` check runs are on
the commit, but the PR's check rollup is `null`, so #82 showed nothing.
- Full evidence is in #85.

## Fix (no PATs, secrets or settings changes)
- **`ci.yml`:** a new `dispatch-status` job runs only on
`workflow_dispatch` (`needs: [check, audit]`, `if: always()`, job-level
`statuses: write`).
- It posts `CI (dispatched) / check` and `CI (dispatched) / audit`
commit statuses on the dispatched SHA, linked to the run. A failure is
posted as `failure`, and a cancelled or skipped job as `error`.
- The names are deliberately different from the required `check` /
`audit`, so a bot-run result can never satisfy branch protection.
  - On PRs and pushes, the job is skipped.
- **`openshell-next.yml`:** the header comment and the #80 comment's
Branch row explain that the result shows as `CI (dispatched)` statuses
and that the PR's own checks wait for approval.
- **Docs:**
- DEVELOPING step 4 covers the statuses, the approval rule, and the
`statuses: write` permission. It also notes that schedules can start
late or be dropped.
- RELEASING promotion step 3 says the required checks come from the PR's
own runs: started by a human push, or by "Approve workflows to run".
  - CHANGELOG has a new line.

## The missed 04:17 UTC schedule
No change needed. GitHub delays or drops scheduled runs under load, and
this repo's only scheduled run so far started 8.5 h late. Daily runs
catch up the next day, or you can dispatch by hand. The docs now say so.

## Not done; options for Ilan
- To have #82's real `check` / `audit` run without a click, the push
would need a GitHub App installation token or a PAT, which means a new
secret. I'm not proposing that for a draft tracking PR.
- No repo-wide setting controls this rule, so no settings change is
needed.

## Verified
- `actionlint` is clean.
- The new step's script and the edited report step pass `bash -n`, and
the status step passes `shellcheck`. I rendered the edited table row to
check it.
- After opening the PR I ran `gh workflow run ci.yml --ref
fix/openshell-next-pr-checks` to check that the statuses appear on this
PR.

Signed-off-by: Ilan Harel <ilan@Ilans-MacBook-Pro.local>
Co-authored-by: Ilan Harel <ilan@Ilans-MacBook-Pro.local>
Signed-off-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Automated by .github/workflows/openshell-next.yml. Upstream changes: NVIDIA/OpenShell@v0.1.8-pre.5...v0.1.8

Signed-off-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Automated by .github/workflows/openshell-next.yml. Upstream changes: NVIDIA/OpenShell@v0.1.8...v0.1.9-pre.1

Signed-off-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Automated by .github/workflows/openshell-next.yml. Upstream changes: NVIDIA/OpenShell@v0.1.9-pre.1...v0.1.9-pre.2

Signed-off-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Automated by .github/workflows/openshell-next.yml. Upstream changes: NVIDIA/OpenShell@v0.1.9-pre.2...v0.1.9-pre.3

Signed-off-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
ilanh added a commit that referenced this pull request Nov 16, 2026
Closes #80

Bumps `main` to the stable [OpenShell
v0.1.8](https://github.com/NVIDIA/OpenShell/releases/tag/v0.1.8)
(released 2026-11-10, tag commit `f3d97659`).

## Change
- `Cargo.toml`/`Cargo.lock`: `openshell-policy-schema` and
`openshell-policy` → `v0.1.8` (via `scripts/openshell-next.sh bump
v0.1.8`).
- Version strings → v0.1.8: generated policy header (`src/generate.rs`,
golden `policy.expected.yaml`, `examples/polyglot` policy),
diff-logs/image notes (`src/*.rs`, three `diff_logs_*` goldens), README,
AGENTS, ROADMAP, Dockerfile, DEVELOPING, `.cargo/audit.toml` comment.
Real v0.1.7 log fixtures and the `upstream-compat.sh` examples keep
their version on purpose.
- `--scaffold-profiles` skeletons explain the v0.1.8 credential
placement rule (#4886); README "Credentials with no provider profile"
too.
- `egress protocol=` help (→ `docs/DIRECTIVES.md` via `gen-docs.sh`):
REST answers WebSocket upgrades with 403 since v0.1.8.
- RELEASING: how to promote when `openshell-next` has already moved past
the stable tag (what happened here).

## v0.1.8 upgrade notes, checked against the release notes and the
v0.1.7→v0.1.8 source
| Upstream change | Effect on annotate | Done here |
|---|---|---|
| #5336 `websocket_credential_rewrite: true` needs `protocol:
websocket`; REST → 403 on upgrade | The gateway validator
(`openshell-policy` `l7_validate.rs`) now rejects it, and `lint
--policy`/`drift`/`generate` run that validator. annotate never emits
the key. | Test
`websocket_credential_rewrite_needs_the_websocket_protocol`; directive
help. Lint warning for WebSocket URLs on `rest` → #95 / PR #97 |
| #5048 OCSF `container.*` → `device.container.*` | `diff-logs` reads no
attribution fields | Unit test
`sandbox_attribution_location_does_not_matter` (both shapes parse the
same); DEVELOPING note. No parser change needed |
| #5278 symlinked fs paths fail launch | Runtime (Landlock) only; not
detectable statically | Documented in PR #97 |
| #5101 `::ffff:a.b.c.d` evaluated as IPv4 | Runtime only, validator
unchanged | Lint warning → #95 / PR #97 |
| #4886 static credentials only in `Authorization` unless placement
declared | Skeletons already emit `auth_style: bearer` + `header_name:
authorization` (explicit) | Skeleton comment + README explain
`header`/`query`/`path` placement |
| Schema / `deny_unknown_fields` | `openshell-policy-schema` diff is one
doc comment; no new or removed keys | — |
| RUSTSEC-2026-0459 | v0.1.8 still pins `sqlx` 0.9.0 / `libsqlite3-sys`
0.37.0 | Ignore stays; comment updated |
| `PROXY_BASELINE_*` | Unchanged in v0.1.8 | `generate --image` baseline
unchanged |
| #4270 FS denial logging | Still open upstream | Notes now say "not
logged by OpenShell 0.1.8" |

## openshell-next / #82
`openshell-next` is already on `v0.1.9-pre.3`. When the workflow merges
this into it, `sync` keeps the newer pin (v0.1.9-pre.3) and the other
files merge cleanly (the branch only differs in
`Cargo.toml`/`Cargo.lock`). After this merges #80 closes, and the
workflow opens a new `openshell-next` tracking issue at the next
upstream tag. **#82** can then either be retitled to track v0.1.9 (it
already carries v0.1.9-pre.3) or be closed and replaced by a fresh
draft; its body/checklist is v0.1.8-specific, so I'd retitle and refresh
it.

## Verified (Rust 1.94.1, Linux)
`scripts/ci-local.sh`: fmt, clippy `-D warnings`, `cargo test --locked`
(all suites, incl. `gateway_policies`), `gen-docs.sh --check`,
`gen-actions-docs.sh --check` all pass.

Not tested: a real v0.1.8 gateway/sandbox run of the playground flow
(#80 checklist item); the iMac/nemoclaw was offline.

Not merging; needs Ilan's approval.

Signed-off-by: Ilan Harel <ilanharel@gmail.com>
Both sides changed the OpenShell pin; kept the newer tag.

Signed-off-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Automated by .github/workflows/openshell-next.yml. Upstream changes: NVIDIA/OpenShell@v0.1.9-pre.3...v0.1.9-pre.4

Signed-off-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Automated by .github/workflows/openshell-next.yml. Upstream changes: NVIDIA/OpenShell@v0.1.9-pre.4...v0.1.9-pre.5

Signed-off-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

openshell-next Tracking the next OpenShell release on the openshell-next branch

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant