You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
0.8.1: Installer repairs your own writable folders and never advises changing system folders #368
Found on 12 Oct: the installer refuses a HOME or data folder under /tmp with advice to run sudo chmod go-w /private/tmp (harmful), and on Ubuntu (umask 0002) it refuses the user's own group-writable ~/.local/share (the printed chmod go-w fix works, but it is an extra step).
Branch fix/install-sticky-tmp (2f9ade572) repairs the user's own writable folders and refuses root- or other-owned writable folders without chmod advice. Cross-family review found that the install-folder parents and the Runtime's directory ancestors are not checked top-down with retained handles, so a shared parent could be swapped between check and use. Not in 0.8.0: develop's behaviour (refuse with a working fix for the user's own folders) stays for the release.
Done means
No message ever advises changing a system or root-owned folder.
The user's own writable folders on the install and data paths are repaired, with every ancestor validated from / down and directory handles kept for later use (installer and Runtime).
Tests: existing and missing folders, links, a replacement race under a shared parent, /tmp refused.
3 November integration. Main reports that PR #523 passed 15 CI jobs with 1 skipped on 2533cc12b. PR #524 passed 14 with 1 skipped on 7656063dc, and its full remote check passed. Main reports that both merges matched their reviewed PR heads. Root checked the local merge objects. 6975b7278 and 046988d51 each have their reviewed PR head as second parent and the exact same source tree as that head. Main still owns the develop push CI result on 046988d51 and the 2.0.1 test-signing dry run under #93. The earlier dispatch failures and remaining installer and release acceptance duties stay open.
2 November after-stable source review: Root found no source blocker in 1ee873aed (main merged back with the exact release tree), 3a2991b8c (previous-release pin now names the verified signed stable 2.0.0 head/release), and 2533cc12b (tag documentation aligns with the source-commit check). The pin’s head CID resolves to the exact previously verified signed stable head. Rename 7656063dc adds exactly those two file changes to reviewed d0ece23bd; independent patch comparison found zero other content changes. This is bounded source GO. Both reported failures in CI 41525149160 remain recorded; new normal-gate/CI results are pending. All nine accepted duties and remaining after-release proof stay OPEN.
2 November stable publication review: Root independently verified the stable 2.0.0 head and release signatures with the pinned maintainer key, exact 7bc3704b5 / tree 794ce317, and fresh public head/release/installer byte parity. Release CID: QmfKYcyaECzckWSHHXbWTdQHc4FNRQXjz2vh3YQ4RCQon8. The public website index matches that source. This verifies publication metadata and the installer; installed journeys and full website/payload checks keep their own proof requirements. Main’s completed verifier reports 10 PASS + VERIFIED. Remaining #368 work follows Anders’s approved after-release order; all nine accepted duties remain OPEN.
Deepen: installed Home CLI directory proof. The retained helper at setup_installed_cli_tests.rs:130 drives the separately selected operator case at :771. Its product Runtime stays bound to the frozen 38-input build; the test-only source map differs at one file. The test uses the real retained Home CLI archive, disposable fixture trust and two isolated prefixes to check refresh, legacy custody and payload-time root replacement.
Home CLI directory refresh now passes native and installed consumer checks. The existing tree refreshes with exact archive members and staged receipts while legacy data stays intact. A data-root replacement during the Carrier response causes refusal and preserves both trees.
Candidate and review: unpublished continuation at d22b33b27de289c33ec6c96462288c28a8c3e36f plus 34 dirty files. PR512 covers the merged parent. The coherent follow-up and normal gate remain pending. Bounded independent reviews checked the source, methods and actual evidence. Required PR and human release approval remain separate.
Native proof: BEFORE compiled, then both intended assertions failed. The first repair compile found a String-to-anyhow conversion error; its original log remains retained. The reviewed conversion produced 22 passing tests, zero failures and zero ignored tests. Default server all-target Clippy passed on that exact source. Earlier failures and review findings remain retained.
Built and installed proof: the frozen product built with the default profile in 51.77s. Built and stable installed SHA-256 both equal 15071a4e (199,197,432 bytes). The same retained source fixture passed five CLI commands with known identity, credentials, 151 UI files, configuration, components, capsule receipt and image stamps unchanged. The canonical receipt binds source commit/tree and binary parity.
Actual directory consumer: the separately armed test compiled in 1m 19s and passed in 14.47s: one selected test, zero failures, zero ignored, 3,313 filtered. Two isolated prefixes ran six actual CLI commands with expected statuses 0,0,0,0,0,1 and exactly six CID requests. The 598,265-byte retained archive supplied all 15 regular members with checked bytes/modes, staged CID/hash receipts and preserved legacy custody. The replacement case preserved both full snapshots. The native renderer stayed unexecuted. This uses disposable canary trust over retained artifact bytes and proves this Mac installer consumer; stable signing/retrust and human acceptance keep their own gates.
Final checks and cleanup: default server all-target Clippy passed in 49.05s. Independent read-only review checked the raw commands, recomputed the CID multiset and inspected retained installed members. All build, upgrade, test and Clippy jobs are terminal; owned children are reaped, Carrier is closed, and known source-Home/IPC/lease open-file checks are clear. The two prefixes and original proof remain retained. Formatting covers the exact reviewed test bytes; diff check passed. Legacy entropy remains absent/unpassed; full normal gate and CI remain pending.
Remaining gate and next owner: Root continues the same guard across other capsule, provider, media, image, on-demand and source-Home effects through activation and cleanup. Safe prefix/profile/resolver/marker/launcher and child propagation; dedicated repair with one privileged consent and decline/no-sudo/failed-repair fallback; honest partial state and same-layout identity replacement refusal; original installation version/channel/signer and reinstall consent; selected-root uninstall; full installed and recovery journeys remain required. Shared parents stay unchanged. Anders's identity-move decision keeps explicit fresh-identity-or-stop choices and retained old folders. Root will finish this coherent scope before committing and running the normal gate. All nine release duties remain open; Anders remains the production signer.
Current integrated-source review: Candidate 7bc3704b5e181e5aa435c2b2cafbd73a996932ef, tree 794ce31737b856f74ec3b6832cf5d231f6aff6b1, changes three packaging and hop-fixture files from 09c25d788. Root found no affected source blocker in the fixed gzip encoding and exact signed-size/hash selection of legacy or new fixture archives. The portability test now pins a worker-independent archive digest. Root ran a diff check and has now read the completed native pre-push log bound to this exact commit/tree. It passed 2,001 JavaScript behavior tests, 132 hop tests, server library tests (3,233 passed, 27 ignored), binary tests (214 passed) and the logged source, formatting, workspace checks and Clippy commands. The log rechecks the source and base, records gate PASS and the normal train fast-forward from 09c. Root has now read completed hosted CI 41375415219 (14 successful jobs, 2 skipped) and package 41375420248 (11 successful jobs), both attempt 1 on this source. Per-step skips are retained. The official Mac and Linux hop receipts pass within their CI-rehearsal scope, with clean owned-process teardown. Six official package-proof inputs have 298 distinct version-and-artifact entries and zero cross-platform conflicts. N and N1 are compared separately, since version-built Runtime bytes differ.
Preparation completed after its first 7bc attempt refused the existing 09c output directory. Root retained both that stop and the successful prepare log. The unsigned signing-input JSON is bound to source 7bc/tree794, package run 41375420248 attempt 1, and SHA-256 31a4fdd95e69153c16a122cc0bc097aa082abab72fdb0f2355a9b48b973c8798. Every official N artifact size/hash agrees with the prepared input (149 unique artifact names; 219 comparisons including shared duplicates). The three component manifests are generated during preparation. Root read hashes and receipts rather than repeating a full payload sweep. Main reports Anders signed and verification passed before canary publication. Root has now read Main's supplied completed verification rerun: ten PASS checks and VERIFIED, including all 152 prepared size/hash records, the exact 155-file signed set and Runtime admission. Its raw log hash is 0334710b; the 155-line signed checksum list hash is 7e6fa0c7. Root read those completed outputs and retained them separately from its independent signature and public-byte checks.
Canary publication: The completed publication log reports 2.0.0-rc.2 published from this source, release CID QmZbQqWFciZ6jyW1mnSNfkBmUaFW9H4TxaPHZhbuyDrksa, head CID Qmea2gPEJXS9LiyywU4qQHj2HPMNpcXTKLFBaBCRFKMcsZ. Root independently verified the signed head and release with the exact candidate installer's verifier and the pinned maintainer DID. Their version, channel, source and hash bindings agree. Fresh public HTTPS head, release and installer bytes match the local signed files. Binary payload downloads, installed journeys and stable acceptance keep their separate proof requirements. Root has now read Main's completed marker logs and their scripts. Four fresh Ubuntu fixtures (both architectures, group-writable and pristine layouts) each passed 13/13 checks; Mac fresh passed 7/7 and Jetson passed 9/9. Linux update/rollback/forward fixtures each passed 18/20, and Mac passed 14/16. The six retained failures check quiet output: the rc.1 executable prints the same INFO diagnostic on each forward update. Exact rc.1 and rc.2 source confirms the change from INFO to debug; these lines add no rc.2 implementation hold.
The update fixtures compare a recorded, nonempty DID and one Local marker hash. Linux Home health is checked after the two forward hops, with rollback version/data checks between them; Mac starts no Home. Jetson's first Home health took 115 seconds, and its file-retention check only confirms a count of at least the three initial files. That count proves neither matching file paths nor contents. Raw Mac fresh output shows the size line before Browser download. These are bounded fixtures; passkey sign-in, desktop/Browser interaction, complete data/recovery journeys and independent installed/served hash bindings remain open.
Mac installed-engine update and recovery (2 November): Main reports that moving the engine aside still left System update failing. Anders then stopped Home and used the CLI update; Main reports rc.2 installed, Home restarted and health passed. Local AI still needs preparation and testing. Root has read the completed clone workaround log, which reaches rc.2 after the engine was moved aside; the real Home's completed CLI, installed-hash and recovery outputs remain author evidence. System acceptance and Local AI restoration stay OPEN.
Root compared the existing official N/N1 package proofs: all three platform engine archive size/hash entries agree between rc.2 and stable. Source supports skipping replacement when the installed component is healthy, its full signed description is unchanged and its capsule record matches. Equal archive hashes alone do not establish all of those conditions or cover interrupted and repair paths. Main's proposed 2.0.1 fix remains separate work with unresolved review findings; its target is not a human waiver. Root has now verified Anders’s originating reply approving stable 2.0.0 from 7bc and package run 41375420248, with the remaining #368 work after 2.0.0. The scoped human ship decision is resolved; the technical evidence and acceptance limits above remain recorded. Main owns execution, and Anders signs.
Both earlier clone runs failed with a local engine installation permission error. Exact 7bc source supports the sealed-tree cleanup diagnosis and wraps local installation errors as Carrier-fetch failures. The original two human failure snapshots were overwritten before preservation; their earlier bindings retain Main's read-output scope, and the retained status/receipt belong to the restarted attempt. All original failures, that correction and the later clone CLI success remain retained. Main owns the exact old-consumer or approved recovery journey, System result, engine restoration and same-Home identity/data proof. All nine duties, full #368 and stable human acceptance remain OPEN.
PR375's first pull-request attempt is separate from the successful push CI: completed attempt 1 has 9 successful jobs, 1 failure and 7 cancellations. Root read GitHub's original failed-check annotation reporting runner disk exhaustion while writing its Worker log; the job-log endpoint returned 404. Main owns its retry and will supply the completed result. Required PR checks and formal human review remain open.
Root reviewed the revised future stable helper. It defaults to a protected, head-matched merge after an exact-source approval. The latest revision also requires a successful command exit on HTTP 200 and distinguishes HTTP 404 from other errors. This resolves the lookup-error hold in source; Main retains the reported operator checks. Main now proposes an explicit human decision on source-tag placement. That decision remains pending. These checks apply to the future stable method. All nine duties, full #368, stable human acceptance and post-publication work stay OPEN.
Root independently checked the 33 common universal archives in the existing Mac, Linux ARM64 and Linux x86-64 inputs. Each compressed file matched its official package-proof size and hash. All decompressed tar hashes agree across the three platforms. The prepare conflict therefore concerns different gzip bytes around equal tar payloads. This check used streams and retained a small measurement receipt. The new stored-block gzip format changes shipped archive bytes and transfer sizes, so the next package needs its own input binding. The first peer review records a checkout mutation and cannot bind its verdict to an exact state; Root's source review binds the Git tree named above. The later peer review binds that same tree.
Predecessor 09c25d7883107318fa97ecedbe4296b05d69e2bf, tree e63c9967023b5ebd0c5f04feaa86ea6ff4547257, retains its completed native gate with 132 passing hop tests and the completed hosted evidence below.
Root read the completed develop CI and package run. CI attempt 1 had 13 successful jobs, one failed JavaScript behavior test and two skipped jobs. The same-source retry passed; the final job set has 14 successes and two skips. The original VZ fixture assertion (did_not_act versus cleanup_pending) remains retained. The unchanged fixture uses short deadlines; timing is Main's assessment, with the exact cause still unproven. Package attempt 1 passed all 11 jobs. Per-step skips remain in the retained API evidence.
Root verified the published digests of the official Linux and Mac hop receipt archives. Both report passed real-Runtime CI rehearsals, clean owned-process teardown and no remaining PIDs. Linux also passed the Ubuntu umask 002 case through authenticated Home, clean Close, update and same-Home preservation. These fixture-trust rehearsals keep their automated scope. Root read the small official package proofs for Mac, Linux x86-64, Linux ARM64 and the inert seed input; they bind this source to the named package artifacts and distinct rc.2/stable version-built inputs. Platform payload downloads, production signatures and stable installed acceptance remain separate.
Earlier failed c11 and 4fa CI attempts, stopped cc7 gate, failed Linux predecessors and incomplete lost-box result remain retained. The Home CLI evidence above keeps its separate unpublished 38-input binding. Profile handling remains the reviewed append-only method with explicit uncertainty on recovery failure; uninstall preserves profiles and names the selected line to remove.
Root read Anders's originating reply approving exact 09c25d7883107318fa97ecedbe4296b05d69e2bf for the described test channel. Anders signs; publication follows successful signed-input verification. This approval covers the current candidate and scope without a repeat question. Stable approval and the #368 deferral remain separate.
Preparation then stopped at 01:12 UTC with conflicting universal asset: artifacts/archive-manager.tar.gz; the retained log also contains an earlier HTTP 502. The completed CI/package passes remain valid within their recorded scope. Main owns the package-combination diagnosis and correction. Root has read no ready-to-sign, signature or publication result for this attempt. Root continues source review only. Full #368, stable/human acceptance and all nine release duties remain OPEN. Future stable execution retains the review-authority and tag-policy holds recorded on #93.
Current owner instruction: Root read Anders's originating reply to the described 7bc rc.2 sequence: he asked Main to continue and rejected repeated blind commit approvals. Main can continue the current release preparation after qualified CI/package and the universal-asset check, then give Anders the signing command. His signing act and successful signed-input verification precede test-channel publication. Routine packaging or fixture corrections within this scope do not need another commit-only question. Stable publication and the full #368 scope retain their separate decisions and acceptance.
Root read the supplied overnight watcher and helper source. The watcher continues into preparation after its checks. Its retry filter selects job names; an actual failed result still needs a reasoned disposition and retained evidence before retry success is accepted. Root queried no current CI status and ran no helper, retry, seed or approval operation.
Acceptance clarification from Anders’s 30 October request, alongside the takeover comment above: check all installation and data-path folders in one pass and report each automatic repair clearly. The later addenda restrict repair to ElastOS-dedicated directories; shared parents and system directories remain unchanged, with a safe dedicated-prefix route when needed. These requirements remain part of 2.0.0. Root retains implementation and qualification ownership; this comment adds acceptance wording, not an implementation or test result.
Scope decision for 2.0.0 (Anders, 31 October 14:31 UTC): no automatic move of an existing identity. Reason given: there are no real users yet. An existing data root under unsafe shared ancestry gets one explicit question: start a new identity in the dedicated ElastOS prefix, with the old folder kept unchanged and unused, or stop. Shared folders are never changed and no chmod advice is printed for them. No later import is promised. Every other requirement in this issue stays required for 2.0.0: own-folder repair only, the dedicated prefix and launcher for unsafe shared parents, the privileged repair consent, and identity, data, signature, channel and downgrade checks.
Safety rule for 2.0.0 (31 October; this is Anders's own 30 October request: refuse only world-writable or foreign-owned folders).
Root cause of the Jetson refusal: the installer (scripts/install.sh around 1219-1266) and the installation writer (check_directory in install_transaction.rs) refuse every group-writable folder. On Ubuntu, home folders are group-writable by default, but each user's group contains only that user, so the refusal is a false alarm on almost every Ubuntu machine. Common installers (rustup, uv, bun, deno) do no permission checks on these folders at all.
Rule: a folder on the install or data path is accepted when it is owned by the user (or by root, for system ancestors) and is not world-writable. Group-writable is accepted. ElastOS creates its own folders and key files owner-only and tightens its own folders if needed; it never changes shared folders.
Only these rare cases differ: a world-writable or other-owned folder on the path leads to the dedicated ElastOS folder (or, for an existing identity there, the fresh-identity-or-stop question); root-owned ElastOS folders from an earlier sudo install get one sudo consent; an odd or newer installed version gets the reinstall offer. Tests: group-writable private Ubuntu folders accepted; world-writable refused; other-owned refused; root-owned ElastOS folder repaired after consent.
Activity
andersalm commentedon Oct 30, 2026
3 November integration. Main reports that PR #523 passed 15 CI jobs with 1 skipped on
2533cc12b. PR #524 passed 14 with 1 skipped on7656063dc, and its full remote check passed. Main reports that both merges matched their reviewed PR heads. Root checked the local merge objects.6975b7278and046988d51each have their reviewed PR head as second parent and the exact same source tree as that head. Main still owns the develop push CI result on046988d51and the 2.0.1 test-signing dry run under #93. The earlier dispatch failures and remaining installer and release acceptance duties stay open.2 November after-stable source review: Root found no source blocker in
1ee873aed(main merged back with the exact release tree),3a2991b8c(previous-release pin now names the verified signed stable 2.0.0 head/release), and2533cc12b(tag documentation aligns with the source-commit check). The pin’s head CID resolves to the exact previously verified signed stable head. Rename7656063dcadds exactly those two file changes to reviewedd0ece23bd; independent patch comparison found zero other content changes. This is bounded source GO. Both reported failures in CI41525149160remain recorded; new normal-gate/CI results are pending. All nine accepted duties and remaining after-release proof stay OPEN.2 November stable publication review: Root independently verified the stable 2.0.0 head and release signatures with the pinned maintainer key, exact
7bc3704b5/ tree794ce317, and fresh public head/release/installer byte parity. Release CID:QmfKYcyaECzckWSHHXbWTdQHc4FNRQXjz2vh3YQ4RCQon8. The public website index matches that source. This verifies publication metadata and the installer; installed journeys and full website/payload checks keep their own proof requirements. Main’s completed verifier reports 10 PASS + VERIFIED. Remaining #368 work follows Anders’s approved after-release order; all nine accepted duties remain OPEN.Deepen: installed Home CLI directory proof. The retained helper at
setup_installed_cli_tests.rs:130drives the separately selected operator case at:771. Its product Runtime stays bound to the frozen 38-input build; the test-only source map differs at one file. The test uses the real retained Home CLI archive, disposable fixture trust and two isolated prefixes to check refresh, legacy custody and payload-time root replacement.Home CLI directory refresh now passes native and installed consumer checks. The existing tree refreshes with exact archive members and staged receipts while legacy data stays intact. A data-root replacement during the Carrier response causes refusal and preserves both trees.
d22b33b27de289c33ec6c96462288c28a8c3e36fplus 34 dirty files. PR512 covers the merged parent. The coherent follow-up and normal gate remain pending. Bounded independent reviews checked the source, methods and actual evidence. Required PR and human release approval remain separate.15071a4e(199,197,432 bytes). The same retained source fixture passed five CLI commands with known identity, credentials, 151 UI files, configuration, components, capsule receipt and image stamps unchanged. The canonical receipt binds source commit/tree and binary parity.0,0,0,0,0,1and exactly six CID requests. The 598,265-byte retained archive supplied all 15 regular members with checked bytes/modes, staged CID/hash receipts and preserved legacy custody. The replacement case preserved both full snapshots. The native renderer stayed unexecuted. This uses disposable canary trust over retained artifact bytes and proves this Mac installer consumer; stable signing/retrust and human acceptance keep their own gates.Current integrated-source review: Candidate
7bc3704b5e181e5aa435c2b2cafbd73a996932ef, tree794ce31737b856f74ec3b6832cf5d231f6aff6b1, changes three packaging and hop-fixture files from09c25d788. Root found no affected source blocker in the fixed gzip encoding and exact signed-size/hash selection of legacy or new fixture archives. The portability test now pins a worker-independent archive digest. Root ran a diff check and has now read the completed native pre-push log bound to this exact commit/tree. It passed 2,001 JavaScript behavior tests, 132 hop tests, server library tests (3,233 passed, 27 ignored), binary tests (214 passed) and the logged source, formatting, workspace checks and Clippy commands. The log rechecks the source and base, records gate PASS and the normal train fast-forward from 09c. Root has now read completed hosted CI 41375415219 (14 successful jobs, 2 skipped) and package 41375420248 (11 successful jobs), both attempt 1 on this source. Per-step skips are retained. The official Mac and Linux hop receipts pass within their CI-rehearsal scope, with clean owned-process teardown. Six official package-proof inputs have 298 distinct version-and-artifact entries and zero cross-platform conflicts. N and N1 are compared separately, since version-built Runtime bytes differ.Preparation completed after its first 7bc attempt refused the existing 09c output directory. Root retained both that stop and the successful prepare log. The unsigned signing-input JSON is bound to source 7bc/tree794, package run 41375420248 attempt 1, and SHA-256
31a4fdd95e69153c16a122cc0bc097aa082abab72fdb0f2355a9b48b973c8798. Every official N artifact size/hash agrees with the prepared input (149 unique artifact names; 219 comparisons including shared duplicates). The three component manifests are generated during preparation. Root read hashes and receipts rather than repeating a full payload sweep. Main reports Anders signed and verification passed before canary publication. Root has now read Main's supplied completed verification rerun: ten PASS checks and VERIFIED, including all 152 prepared size/hash records, the exact 155-file signed set and Runtime admission. Its raw log hash is0334710b; the 155-line signed checksum list hash is7e6fa0c7. Root read those completed outputs and retained them separately from its independent signature and public-byte checks.Canary publication: The completed publication log reports 2.0.0-rc.2 published from this source, release CID
QmZbQqWFciZ6jyW1mnSNfkBmUaFW9H4TxaPHZhbuyDrksa, head CIDQmea2gPEJXS9LiyywU4qQHj2HPMNpcXTKLFBaBCRFKMcsZ. Root independently verified the signed head and release with the exact candidate installer's verifier and the pinned maintainer DID. Their version, channel, source and hash bindings agree. Fresh public HTTPS head, release and installer bytes match the local signed files. Binary payload downloads, installed journeys and stable acceptance keep their separate proof requirements. Root has now read Main's completed marker logs and their scripts. Four fresh Ubuntu fixtures (both architectures, group-writable and pristine layouts) each passed 13/13 checks; Mac fresh passed 7/7 and Jetson passed 9/9. Linux update/rollback/forward fixtures each passed 18/20, and Mac passed 14/16. The six retained failures check quiet output: the rc.1 executable prints the same INFO diagnostic on each forward update. Exact rc.1 and rc.2 source confirms the change from INFO to debug; these lines add no rc.2 implementation hold.The update fixtures compare a recorded, nonempty DID and one Local marker hash. Linux Home health is checked after the two forward hops, with rollback version/data checks between them; Mac starts no Home. Jetson's first Home health took 115 seconds, and its file-retention check only confirms a count of at least the three initial files. That count proves neither matching file paths nor contents. Raw Mac fresh output shows the size line before Browser download. These are bounded fixtures; passkey sign-in, desktop/Browser interaction, complete data/recovery journeys and independent installed/served hash bindings remain open.
Mac installed-engine update and recovery (2 November): Main reports that moving the engine aside still left System update failing. Anders then stopped Home and used the CLI update; Main reports rc.2 installed, Home restarted and health passed. Local AI still needs preparation and testing. Root has read the completed clone workaround log, which reaches rc.2 after the engine was moved aside; the real Home's completed CLI, installed-hash and recovery outputs remain author evidence. System acceptance and Local AI restoration stay OPEN.
Root compared the existing official N/N1 package proofs: all three platform engine archive size/hash entries agree between rc.2 and stable. Source supports skipping replacement when the installed component is healthy, its full signed description is unchanged and its capsule record matches. Equal archive hashes alone do not establish all of those conditions or cover interrupted and repair paths. Main's proposed 2.0.1 fix remains separate work with unresolved review findings; its target is not a human waiver. Root has now verified Anders’s originating reply approving stable 2.0.0 from 7bc and package run 41375420248, with the remaining #368 work after 2.0.0. The scoped human ship decision is resolved; the technical evidence and acceptance limits above remain recorded. Main owns execution, and Anders signs.
Both earlier clone runs failed with a local engine installation permission error. Exact 7bc source supports the sealed-tree cleanup diagnosis and wraps local installation errors as Carrier-fetch failures. The original two human failure snapshots were overwritten before preservation; their earlier bindings retain Main's read-output scope, and the retained status/receipt belong to the restarted attempt. All original failures, that correction and the later clone CLI success remain retained. Main owns the exact old-consumer or approved recovery journey, System result, engine restoration and same-Home identity/data proof. All nine duties, full #368 and stable human acceptance remain OPEN.
PR375's first pull-request attempt is separate from the successful push CI: completed attempt 1 has 9 successful jobs, 1 failure and 7 cancellations. Root read GitHub's original failed-check annotation reporting runner disk exhaustion while writing its Worker log; the job-log endpoint returned 404. Main owns its retry and will supply the completed result. Required PR checks and formal human review remain open.
Root reviewed the revised future stable helper. It defaults to a protected, head-matched merge after an exact-source approval. The latest revision also requires a successful command exit on HTTP 200 and distinguishes HTTP 404 from other errors. This resolves the lookup-error hold in source; Main retains the reported operator checks. Main now proposes an explicit human decision on source-tag placement. That decision remains pending. These checks apply to the future stable method. All nine duties, full #368, stable human acceptance and post-publication work stay OPEN.
Root independently checked the 33 common universal archives in the existing Mac, Linux ARM64 and Linux x86-64 inputs. Each compressed file matched its official package-proof size and hash. All decompressed tar hashes agree across the three platforms. The prepare conflict therefore concerns different gzip bytes around equal tar payloads. This check used streams and retained a small measurement receipt. The new stored-block gzip format changes shipped archive bytes and transfer sizes, so the next package needs its own input binding. The first peer review records a checkout mutation and cannot bind its verdict to an exact state; Root's source review binds the Git tree named above. The later peer review binds that same tree.
Predecessor
09c25d7883107318fa97ecedbe4296b05d69e2bf, treee63c9967023b5ebd0c5f04feaa86ea6ff4547257, retains its completed native gate with 132 passing hop tests and the completed hosted evidence below.Root read the completed develop CI and package run. CI attempt 1 had 13 successful jobs, one failed JavaScript behavior test and two skipped jobs. The same-source retry passed; the final job set has 14 successes and two skips. The original VZ fixture assertion (
did_not_actversuscleanup_pending) remains retained. The unchanged fixture uses short deadlines; timing is Main's assessment, with the exact cause still unproven. Package attempt 1 passed all 11 jobs. Per-step skips remain in the retained API evidence.Root verified the published digests of the official Linux and Mac hop receipt archives. Both report passed real-Runtime CI rehearsals, clean owned-process teardown and no remaining PIDs. Linux also passed the Ubuntu umask 002 case through authenticated Home, clean Close, update and same-Home preservation. These fixture-trust rehearsals keep their automated scope. Root read the small official package proofs for Mac, Linux x86-64, Linux ARM64 and the inert seed input; they bind this source to the named package artifacts and distinct rc.2/stable version-built inputs. Platform payload downloads, production signatures and stable installed acceptance remain separate.
Earlier failed c11 and 4fa CI attempts, stopped cc7 gate, failed Linux predecessors and incomplete lost-box result remain retained. The Home CLI evidence above keeps its separate unpublished 38-input binding. Profile handling remains the reviewed append-only method with explicit uncertainty on recovery failure; uninstall preserves profiles and names the selected line to remove.
Root read Anders's originating reply approving exact
09c25d7883107318fa97ecedbe4296b05d69e2bffor the described test channel. Anders signs; publication follows successful signed-input verification. This approval covers the current candidate and scope without a repeat question. Stable approval and the #368 deferral remain separate.Preparation then stopped at 01:12 UTC with
conflicting universal asset: artifacts/archive-manager.tar.gz; the retained log also contains an earlier HTTP 502. The completed CI/package passes remain valid within their recorded scope. Main owns the package-combination diagnosis and correction. Root has read no ready-to-sign, signature or publication result for this attempt. Root continues source review only. Full #368, stable/human acceptance and all nine release duties remain OPEN. Future stable execution retains the review-authority and tag-policy holds recorded on #93.Current owner instruction: Root read Anders's originating reply to the described 7bc rc.2 sequence: he asked Main to continue and rejected repeated blind commit approvals. Main can continue the current release preparation after qualified CI/package and the universal-asset check, then give Anders the signing command. His signing act and successful signed-input verification precede test-channel publication. Routine packaging or fixture corrections within this scope do not need another commit-only question. Stable publication and the full #368 scope retain their separate decisions and acceptance.
Root read the supplied overnight watcher and helper source. The watcher continues into preparation after its checks. Its retry filter selects job names; an actual failed result still needs a reasoned disposition and retained evidence before retry success is accepted. Root queried no current CI status and ran no helper, retry, seed or approval operation.
andersalm commentedon Oct 31, 2026
Acceptance clarification from Anders’s 30 October request, alongside the takeover comment above: check all installation and data-path folders in one pass and report each automatic repair clearly. The later addenda restrict repair to ElastOS-dedicated directories; shared parents and system directories remain unchanged, with a safe dedicated-prefix route when needed. These requirements remain part of 2.0.0. Root retains implementation and qualification ownership; this comment adds acceptance wording, not an implementation or test result.
andersalm commentedon Oct 31, 2026
Scope decision for 2.0.0 (Anders, 31 October 14:31 UTC): no automatic move of an existing identity. Reason given: there are no real users yet. An existing data root under unsafe shared ancestry gets one explicit question: start a new identity in the dedicated ElastOS prefix, with the old folder kept unchanged and unused, or stop. Shared folders are never changed and no chmod advice is printed for them. No later import is promised. Every other requirement in this issue stays required for 2.0.0: own-folder repair only, the dedicated prefix and launcher for unsafe shared parents, the privileged repair consent, and identity, data, signature, channel and downgrade checks.
andersalm commentedon Oct 31, 2026
Safety rule for 2.0.0 (31 October; this is Anders's own 30 October request: refuse only world-writable or foreign-owned folders).
Root cause of the Jetson refusal: the installer (scripts/install.sh around 1219-1266) and the installation writer (check_directory in install_transaction.rs) refuse every group-writable folder. On Ubuntu, home folders are group-writable by default, but each user's group contains only that user, so the refusal is a false alarm on almost every Ubuntu machine. Common installers (rustup, uv, bun, deno) do no permission checks on these folders at all.
Rule: a folder on the install or data path is accepted when it is owned by the user (or by root, for system ancestors) and is not world-writable. Group-writable is accepted. ElastOS creates its own folders and key files owner-only and tightens its own folders if needed; it never changes shared folders.
Only these rare cases differ: a world-writable or other-owned folder on the path leads to the dedicated ElastOS folder (or, for an existing identity there, the fresh-identity-or-stop question); root-owned ElastOS folders from an earlier sudo install get one sudo consent; an odd or newer installed version gets the reinstall offer. Tests: group-writable private Ubuntu folders accepted; world-writable refused; other-owned refused; root-owned ElastOS folder repaired after consent.
Accept group-writable owned data roots and shared ancestors on first …
CI: Linux update hop installs on the default Ubuntu account layout (#368
Custody and IPFS walks relax group write only for the data root and a…
CI hop Ubuntu case: child-only umask 002, same-Home preservation, typ…
Custody and IPFS walks allow the sticky-folder exception only at DATA…