chore(hygiene): Node 24 actions, --version flag, update-apply install priority#4
Merged
Conversation
… priority
Three distribution/operability fixes batched ahead of the June 16
Node 24 forcing date:
- Workflows: checkout v4→v5, cache v4→v5, upload-artifact v4→v6,
download-artifact v4→v7, softprops/action-gh-release v2→v3. All are
the first majors that RUN on node24 (upload v5 / download v6 only
carried preliminary support while defaulting to node20; gh-release
v3 is a runtime-only major with no API change). GitHub forces
node24 on June 16, 2026; node20 is removed from runners Sep 16.
- CLI: #[command(version)] — s5d 0.11.0-alpha.8 now answers from
CARGO_PKG_VERSION. Its absence meant installed-binary versions
could only be guessed from behavior probes (bit us twice during
release verification).
- update apply: install_binary_from_repo preferred the tracked
prebuilt over building from source, silently installing whatever
binary was last committed (stale at both alpha.7 and alpha.8).
Inverted: cargo build from the checked-out revision is the default;
the prebuilt is a loud-warning fallback for hosts without a Rust
toolchain. The install itself was already inode-safe (tmp + rename),
so the macOS in-place-cp signature kill cannot happen through the
canonical path — only through manual cp, documented in RAN-480.
- bin/s5d (6.9M Mach-O, stale since pre-alpha.7) deleted: nothing
reads it — prebuilt_binary() only resolves bin/s5d-{os}-{arch} and
install.sh uses the same pattern. bin/s5d-darwin-arm64 rebuilt at
alpha.8 so the toolchain-less fallback matches the release again.
…r; upload v7 PR #4 tribunal round (both auditors CONCERNS/MEDIUM, disjoint findings): - Codex: the fallback fired on BUILD FAILURE too, not just missing cargo — broken source + existing prebuilt would silently install a lagging binary under an overall success message, recreating exactly the stale-install failure mode this change set out to kill. Now a failed build on a host with cargo is a hard error; the prebuilt fallback is reachable only when cargo cannot be spawned. - Gemini: the hardcoded rust/target/release path breaks under CARGO_TARGET_DIR / .cargo config redirects (build succeeds, install reads a stale or missing path). Pinned with explicit --target-dir. - Gemini flagged upload v6 / download v7 as a mismatch; Codex verified compatibility (shared v4+ artifact backend, download v7 rejects only v3-and-older). Mechanism beats pattern, but synchronizing to v7/v7 costs one line and removes the question permanently — bumped.
random1st
added a commit
that referenced
this pull request
Jun 11, 2026
Findings from an end-to-end cold-install audit (fresh public clone, isolated HOME, full lifecycle probe — install → init → spec lifecycle → ci init/exec all green after these fixes): - install.sh preferred the tracked prebuilt over building from the checkout — the same silent-stale trap fixed in update apply (PR #4) lived on in the installer. Inverted: cargo build first, prebuilt is a warned fallback for toolchain-less hosts. rm before cp so the installed binary always gets a fresh inode (macOS AMFI; the codesign re-sign step already present stays as belt-and-suspenders). - Plugin hooks failed with exit 127 on every session/tool-use for a marketplace install without the CLI binary. Every hook command is now guarded with an if-form "command -v s5d" check: absent binary means exit 0 (SessionStart prints a one-line install hint instead), and — deliberately if-form, not "&& ... || true" — a present binary's exit code passes through untouched so blocking enforcement hooks still block (verified both ways with a fake exit-2 binary). - All three distribution manifests (.claude-plugin, .codex-plugin, gemini-extension) still claimed version 0.3.0 against an 0.11.x binary. Synced to 0.11.0-alpha.9 and added the four version touchpoints to the release agent so the next release cannot ship stale manifests. update apply itself audited clean: fetch+pull --ff-only before rebuild, tmp+rename install (inode-safe). 193 tests green.
random1st
added a commit
that referenced
this pull request
Jun 11, 2026
Findings from an end-to-end cold-install audit (fresh public clone, isolated HOME, full lifecycle probe — install → init → spec lifecycle → ci init/exec all green after these fixes): - install.sh preferred the tracked prebuilt over building from the checkout — the same silent-stale trap fixed in update apply (PR #4) lived on in the installer. Inverted: cargo build first, prebuilt is a warned fallback for toolchain-less hosts. rm before cp so the installed binary always gets a fresh inode (macOS AMFI; the codesign re-sign step already present stays as belt-and-suspenders). - Plugin hooks failed with exit 127 on every session/tool-use for a marketplace install without the CLI binary. Every hook command is now guarded with an if-form "command -v s5d" check: absent binary means exit 0 (SessionStart prints a one-line install hint instead), and — deliberately if-form, not "&& ... || true" — a present binary's exit code passes through untouched so blocking enforcement hooks still block (verified both ways with a fake exit-2 binary). - All three distribution manifests (.claude-plugin, .codex-plugin, gemini-extension) still claimed version 0.3.0 against an 0.11.x binary. Synced to 0.11.0-alpha.9 and added the four version touchpoints to the release agent so the next release cannot ship stale manifests. update apply itself audited clean: fetch+pull --ff-only before rebuild, tmp+rename install (inode-safe). 193 tests green.
random1st
added a commit
that referenced
this pull request
Jul 1, 2026
…osed Scenario sweep (RAN-481, beta bar = 15/15): every previously uncovered or partial scenario now has a pinning test. The sweep was empirical — each gap was probed against the live binary first, and three of the six probes hit real defects, which is the point of the exercise. Closed scenarios: - #12 MCP containment: new tests/mcp_wire.rs speaks line-delimited JSON-RPC to the spawned binary — initialize/tools-list roundtrip, garbage/notification/unknown-method survival, and a full lifecycle over MCP proving the placeholder-import guard fires on this path too. - #9 reconcile semantics: reconcile is verify+stamp with fail-closed refusal — restore of uuid-bearing drift is structurally impossible (resolve() mints fresh uuids), so the test pins what matters: a failed reconcile never re-baselines the ledger and drift stays visible afterwards. - #14 partial corruption: re-init now restores deleted core files. - #5 decision tier: clean end-to-end without --force; the canonical path is >=3 hypotheses, evidence on each, acceptance criteria on the problem card, winner spec_ref via s5d new --hypothesis-id, and an explicit challenge summary. - #4 high assurance: full flow with configured test+contract commands on top of schema+graph+review, no --force. - #11 show: smoke assertions inside the #4/#5 flows. Bugs fixed (each found by its scenario probe): 1. Bulk 's5d reconcile' exited 0 while individual reconciles failed — an agent reading the exit code saw success over persistent drift. Now counts failures, prints them, and bails non-zero. 2. Re-running init did not recreate deleted config/ledger/index.yaml (only subdirs were repaired) — a deleted file bricked the project. ensure_core_files restores missing files with defaults and reports them; existing files, even corrupted ones, are never overwritten because that is a data-loss decision that stays with the user. 3. add-hypothesis/add-evidence rejected high-tier specs while the high scaffold ships a review gate satisfiable ONLY by gate:review evidence on a hypothesis — the canonical path of the most safety-critical tier was unpassable without --force. Loosened to decision|high at all four sites (CLI+MCP); decide stays decision-only since choosing a winner is decision-tier semantics. Known limitation documented, deliberately not fixed: owning_package is outside compute_state_fingerprint (G:type:id:uuid), so its corruption is invisible to drift-check even though rollback ownership depends on it. Pinned by a documenting test; the fix changes the fingerprint and would invalidate every existing baseline, so it needs an explicit migration decision. README tier table corrected: Decision and High carry a Review default gate (the table claimed None / Schema+Graph; code says otherwise).
random1st
added a commit
that referenced
this pull request
Jul 1, 2026
…r; upload v7 PR #4 tribunal round (both auditors CONCERNS/MEDIUM, disjoint findings): - Codex: the fallback fired on BUILD FAILURE too, not just missing cargo — broken source + existing prebuilt would silently install a lagging binary under an overall success message, recreating exactly the stale-install failure mode this change set out to kill. Now a failed build on a host with cargo is a hard error; the prebuilt fallback is reachable only when cargo cannot be spawned. - Gemini: the hardcoded rust/target/release path breaks under CARGO_TARGET_DIR / .cargo config redirects (build succeeds, install reads a stale or missing path). Pinned with explicit --target-dir. - Gemini flagged upload v6 / download v7 as a mismatch; Codex verified compatibility (shared v4+ artifact backend, download v7 rejects only v3-and-older). Mechanism beats pattern, but synchronizing to v7/v7 costs one line and removes the question permanently — bumped.
random1st
added a commit
that referenced
this pull request
Jul 1, 2026
Findings from an end-to-end cold-install audit (fresh public clone, isolated HOME, full lifecycle probe — install → init → spec lifecycle → ci init/exec all green after these fixes): - install.sh preferred the tracked prebuilt over building from the checkout — the same silent-stale trap fixed in update apply (PR #4) lived on in the installer. Inverted: cargo build first, prebuilt is a warned fallback for toolchain-less hosts. rm before cp so the installed binary always gets a fresh inode (macOS AMFI; the codesign re-sign step already present stays as belt-and-suspenders). - Plugin hooks failed with exit 127 on every session/tool-use for a marketplace install without the CLI binary. Every hook command is now guarded with an if-form "command -v s5d" check: absent binary means exit 0 (SessionStart prints a one-line install hint instead), and — deliberately if-form, not "&& ... || true" — a present binary's exit code passes through untouched so blocking enforcement hooks still block (verified both ways with a fake exit-2 binary). - All three distribution manifests (.claude-plugin, .codex-plugin, gemini-extension) still claimed version 0.3.0 against an 0.11.x binary. Synced to 0.11.0-alpha.9 and added the four version touchpoints to the release agent so the next release cannot ship stale manifests. update apply itself audited clean: fetch+pull --ff-only before rebuild, tmp+rename install (inode-safe). 193 tests green.
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
Distribution/operability hygiene ahead of the June 16 Node 24 forcing date:
s5d --version: answers from CARGO_PKG_VERSION. Installed-binary version could previously only be inferred from behavior probes — bit us during release verification twice.update applypriority inversion: cargo build from the checked-out revision is now the default install source; the tracked prebuilt (stale at both alpha.7 and alpha.8) is a loud-warning fallback for hosts without a toolchain. Install path was already inode-safe (tmp + rename).bin/s5ddeleted (6.9M stale Mach-O nothing references;prebuilt_binary()and install.sh both resolve onlys5d-{os}-{arch}).bin/s5d-darwin-arm64rebuilt at alpha.8.Tests: 188 passed, fmt/clippy clean,
s5d --version→s5d 0.11.0-alpha.8verified on the rebuilt prebuilt and the installed binary.Sources for the action versions: Node 20 deprecation changelog, checkout v5 (node24), upload-artifact releases, download-artifact releases, gh-release v3 (node24).
🤖 Generated with Claude Code