Session continuity / FEATURE

Restart recovery

Recover pending work or interaction state after restart; state exactly what survives.

3connected products
Explore in graph ↗

Standing condition: these are documentation claims at the source capture date, with outcome “claimed”; no runtime execution is recorded. Dossier applicability and limitations apply throughout. Expand a row for its statement, specific scope and contradictions. Numbered citations resolve to the source locator and capture date below. Missing evidence means Unknown, not No. Sources are unpinned unless marked “Pinned”; this marks source stability, not stronger support or runtime verification. Reuse counts refer to claims in the full dossier; repeated citations do not represent independent sources.

Yes Documented within scope Partial Limited form of the feature No Explicit evidence of absence Unknown No evidence recorded. These are support assessments, not quality scores or runtime verification.

3 products have Yes or Partial documentation. Review freshness and attestation counts describe the dossier, not runtime quality. Gaps below are product-wide, not necessarily specific to this feature.

Nimbalyst →

Needs recheck · 2026-10-130 attestations · rev 2026-09-13.4 · unreviewed
Restart recoveryPartial
Explore Restart recovery →

Scheduled session wakeups survive app restarts.

Scope: Pending wakeups only; the app must be running to fire. Not arbitrary process recovery.

3 open gaps · applicability & verification plan

What we do not know yet about Nimbalyst

  • Measure how agents coordinate when tasks touch the same files.
  • Test recovery and session continuity across desktop and mobile.
  • Inventory leads for native cross-session messaging, live context meters and internal orchestration APIs still need matching public source evidence. Claude Agent team behavior must be checked against the installed runtime version.

Applicability

Unpinned public documentation as of 2026-09-13

Documentation review only

Excludes: Runtime behavior; Performance comparisons; Security guarantees

Recheck when

  • Documentation changes
  • New product release
  • License change

Needs recheck · due

Verification plan

  1. Pin a product release or repository commit and record the host environment.
  2. Create a disposable repository with a passing baseline test.
  3. Run a bounded change with two agents where supported; record permissions and logs.
  4. Record a failed attempt, actual outcome, and artifacts without secrets.
  5. Have another person reproduce the result before labeling it reproduced.

Provenance & review

Testedknowhow initial research

Commissioned by a Nimbalyst maintainer; not an independent comparison.

Revision 2026-09-13.4 · unreviewed · checked

0 attestations · Content: Unlicensed prototype; redistribution decision pending

Sources · 1

Conductor →

Needs recheck · 2026-10-130 attestations · rev 2026-09-13.4 · unreviewed

Unknown — no evidence recorded

2 open gaps · applicability & verification plan

What we do not know yet about Conductor

  • Compare cloud sandbox isolation with local worktree isolation.
  • Test shared workspace permissions and simultaneous edits.

Applicability

Unpinned public documentation as of 2026-09-13

Documentation review only

Excludes: Runtime behavior; Performance comparisons; Security guarantees

Recheck when

  • Documentation changes
  • New product release
  • License change

Needs recheck · due

Verification plan

  1. Pin a product release or repository commit and record the host environment.
  2. Create a disposable repository with a passing baseline test.
  3. Run a bounded change with two agents where supported; record permissions and logs.
  4. Record a failed attempt, actual outcome, and artifacts without secrets.
  5. Have another person reproduce the result before labeling it reproduced.

Provenance & review

Testedknowhow initial research

Commissioned by a Nimbalyst maintainer; not an independent comparison.

Revision 2026-09-13.4 · unreviewed · checked

0 attestations · Content: Unlicensed prototype; redistribution decision pending

Superset →

Needs recheck · 2026-10-130 attestations · rev 2026-09-13.4 · unreviewed

Unknown — no evidence recorded

2 open gaps · applicability & verification plan

What we do not know yet about Superset

  • Cloud offering is marked as coming; distinguish it from existing remote hosts.
  • Inspect ELv2 restrictions before planning redistribution or hosted reuse.

Applicability

Unpinned public documentation as of 2026-09-13

Documentation review only

Excludes: Runtime behavior; Performance comparisons; Security guarantees

Recheck when

  • Documentation changes
  • New product release
  • License change

Needs recheck · due

Verification plan

  1. Pin a product release or repository commit and record the host environment.
  2. Create a disposable repository with a passing baseline test.
  3. Run a bounded change with two agents where supported; record permissions and logs.
  4. Record a failed attempt, actual outcome, and artifacts without secrets.
  5. Have another person reproduce the result before labeling it reproduced.

Provenance & review

Testedknowhow initial research

Commissioned by a Nimbalyst maintainer; not an independent comparison.

Revision 2026-09-13.4 · unreviewed · checked

0 attestations · Content: Unlicensed prototype; redistribution decision pending

Vibe Kanban →

Needs recheck · 2026-10-130 attestations · rev 2026-09-13.4 · unreviewed

Unknown — no evidence recorded

2 open gaps · applicability & verification plan

What we do not know yet about Vibe Kanban

  • Project announces sunsetting and community maintenance; evaluate maintenance risk.
  • Recheck the supported release and setup before attempting reproduction.

Applicability

Unpinned public documentation as of 2026-09-13

Documentation review only

Excludes: Runtime behavior; Performance comparisons; Security guarantees

Recheck when

  • Documentation changes
  • New product release
  • License change

Needs recheck · due

Verification plan

  1. Pin a product release or repository commit and record the host environment.
  2. Create a disposable repository with a passing baseline test.
  3. Run a bounded change with two agents where supported; record permissions and logs.
  4. Record a failed attempt, actual outcome, and artifacts without secrets.
  5. Have another person reproduce the result before labeling it reproduced.

Provenance & review

Testedknowhow initial research

Commissioned by a Nimbalyst maintainer; not an independent comparison.

Revision 2026-09-13.4 · unreviewed · checked

0 attestations · Content: Unlicensed prototype; redistribution decision pending

Cursor →

Needs recheck · 2026-10-130 attestations · rev 2026-09-13.4 · unreviewed

Unknown — no evidence recorded

2 open gaps · applicability & verification plan

What we do not know yet about Cursor

  • Test reproducible environment setup and artifacts for cloud tasks.
  • Differentiate model choice from support for external agent runtimes.

Applicability

Unpinned public documentation as of 2026-09-13

Documentation review only

Excludes: Runtime behavior; Performance comparisons; Security guarantees

Recheck when

  • Documentation changes
  • New product release
  • License change

Needs recheck · due

Verification plan

  1. Pin a product release or repository commit and record the host environment.
  2. Create a disposable repository with a passing baseline test.
  3. Run a bounded change with two agents where supported; record permissions and logs.
  4. Record a failed attempt, actual outcome, and artifacts without secrets.
  5. Have another person reproduce the result before labeling it reproduced.

Provenance & review

Testedknowhow initial research

Commissioned by a Nimbalyst maintainer; not an independent comparison.

Revision 2026-09-13.4 · unreviewed · checked

0 attestations · Content: Unlicensed prototype; redistribution decision pending

Devin →

Needs recheck · 2026-10-130 attestations · rev 2026-09-13.4 · unreviewed

Unknown — no evidence recorded

2 open gaps · applicability & verification plan

What we do not know yet about Devin

  • Measure intervention and success on a bounded task, rather than relying on vendor efficacy claims.
  • Check task-specific permissions and cloud environment constraints.

Applicability

Unpinned public documentation as of 2026-09-13

Documentation review only

Excludes: Runtime behavior; Performance comparisons; Security guarantees

Recheck when

  • Documentation changes
  • New product release
  • License change

Needs recheck · due

Verification plan

  1. Pin a product release or repository commit and record the host environment.
  2. Create a disposable repository with a passing baseline test.
  3. Run a bounded change with two agents where supported; record permissions and logs.
  4. Record a failed attempt, actual outcome, and artifacts without secrets.
  5. Have another person reproduce the result before labeling it reproduced.

Provenance & review

Testedknowhow initial research

Commissioned by a Nimbalyst maintainer; not an independent comparison.

Revision 2026-09-13.4 · unreviewed · checked

0 attestations · Content: Unlicensed prototype; redistribution decision pending

OpenHands →

Needs recheck · 2026-10-130 attestations · rev 2026-09-13.4 · unreviewed

Unknown — no evidence recorded

2 open gaps · applicability & verification plan

What we do not know yet about OpenHands

  • Pin the Agent Canvas revision before following older OpenHands setup guides.
  • Verify isolation and credentials for the chosen backend.

Applicability

Unpinned public documentation as of 2026-09-13

Documentation review only

Excludes: Runtime behavior; Performance comparisons; Security guarantees

Recheck when

  • Documentation changes
  • New product release
  • License change

Needs recheck · due

Verification plan

  1. Pin a product release or repository commit and record the host environment.
  2. Create a disposable repository with a passing baseline test.
  3. Run a bounded change with two agents where supported; record permissions and logs.
  4. Record a failed attempt, actual outcome, and artifacts without secrets.
  5. Have another person reproduce the result before labeling it reproduced.

Provenance & review

Testedknowhow initial research

Commissioned by a Nimbalyst maintainer; not an independent comparison.

Revision 2026-09-13.4 · unreviewed · checked

0 attestations · Content: Unlicensed prototype; redistribution decision pending

Emdash →

Needs recheck · 2026-10-130 attestations · rev 2026-09-13.4 · unreviewed

Unknown — no evidence recorded

2 open gaps · applicability & verification plan

What we do not know yet about Emdash

  • Remote execution requires your own configured host.
  • Session hooks and resumption depend on the agent provider.

Applicability

Repository documentation at fccf350847097c8f9d9de3e7e5c98b0d292a4554

Documentation review only

Excludes: Runtime behavior; Performance comparisons; Security guarantees

Recheck when

  • Documentation changes
  • New product release
  • License change

Needs recheck · due

Verification plan

  1. Pin a product release or repository commit and record the host environment.
  2. Create a disposable repository with a passing baseline test.
  3. Run a bounded change with two agents where supported; record permissions and logs.
  4. Record a failed attempt, actual outcome, and artifacts without secrets.
  5. Have another person reproduce the result before labeling it reproduced.

Provenance & review

Testedknowhow initial research

Commissioned by a Nimbalyst maintainer; not an independent comparison.

Revision 2026-09-13.4 · unreviewed · checked

0 attestations · Content: Unlicensed prototype; redistribution decision pending

Claude Squad →

Needs recheck · 2026-10-130 attestations · rev 2026-09-13.4 · unreviewed

Unknown — no evidence recorded

2 open gaps · applicability & verification plan

What we do not know yet about Claude Squad

  • Requires tmux and the GitHub CLI.
  • Auto-accept exists; do not mistake it for safe sandboxing.

Applicability

Repository documentation at ce1ffb4392b01f38e2c4599c7c84d2a93973b138

Documentation review only

Excludes: Runtime behavior; Performance comparisons; Security guarantees

Recheck when

  • Documentation changes
  • New product release
  • License change

Needs recheck · due

Verification plan

  1. Pin a product release or repository commit and record the host environment.
  2. Create a disposable repository with a passing baseline test.
  3. Run a bounded change with two agents where supported; record permissions and logs.
  4. Record a failed attempt, actual outcome, and artifacts without secrets.
  5. Have another person reproduce the result before labeling it reproduced.

Provenance & review

Testedknowhow initial research

Commissioned by a Nimbalyst maintainer; not an independent comparison.

Revision 2026-09-13.4 · unreviewed · checked

0 attestations · Content: Unlicensed prototype; redistribution decision pending

CloudCLI →

Needs recheck · 2026-10-130 attestations · rev 2026-09-13.4 · unreviewed

Unknown — no evidence recorded

2 open gaps · applicability & verification plan

What we do not know yet about CloudCLI

  • Task planning requires the optional TaskMaster integration.
  • Docker sandbox support is experimental; hosted and self-hosted surfaces differ.

Applicability

Repository documentation at 5e73a49b89b4f13766fc2e22723297a36e7dcef2; Unpinned public documentation and/or pinned README sections reviewed 2026-09-13; per-claim source scope applies.

Documentation review only

Excludes: Runtime behavior; Performance comparisons; Security guarantees

Recheck when

  • Documentation changes
  • New product release
  • License change

Needs recheck · due

Verification plan

  1. Pin a product release or repository commit and record the host environment.
  2. Create a disposable repository with a passing baseline test.
  3. Run a bounded change with two agents where supported; record permissions and logs.
  4. Record a failed attempt, actual outcome, and artifacts without secrets.
  5. Have another person reproduce the result before labeling it reproduced.

Provenance & review

Testedknowhow initial research

Commissioned by a Nimbalyst maintainer; not an independent comparison.

Revision 2026-09-13.4 · unreviewed · checked

0 attestations · Content: Unlicensed prototype; redistribution decision pending

Opcode →

Needs recheck · 2026-10-130 attestations · rev 2026-09-13.4 · unreviewed

Unknown — no evidence recorded

2 open gaps · applicability & verification plan

What we do not know yet about Opcode

  • README says release executables will be published later; source-build feasibility needs testing.
  • Separate agent processes do not establish filesystem isolation.

Applicability

Repository documentation at 70c16d8a4910db48cd9684aeacdd431caefd7d71; Unpinned public documentation and/or pinned README sections reviewed 2026-09-13; per-claim source scope applies.

Documentation review only

Excludes: Runtime behavior; Performance comparisons; Security guarantees

Recheck when

  • Documentation changes
  • New product release
  • License change

Needs recheck · due

Verification plan

  1. Pin a product release or repository commit and record the host environment.
  2. Create a disposable repository with a passing baseline test.
  3. Run a bounded change with two agents where supported; record permissions and logs.
  4. Record a failed attempt, actual outcome, and artifacts without secrets.
  5. Have another person reproduce the result before labeling it reproduced.

Provenance & review

Testedknowhow initial research

Commissioned by a Nimbalyst maintainer; not an independent comparison.

Revision 2026-09-13.4 · unreviewed · checked

0 attestations · Content: Unlicensed prototype; redistribution decision pending

CodexMonitor →

Needs recheck · 2026-10-130 attestations · rev 2026-09-13.4 · unreviewed

Unknown — no evidence recorded

2 open gaps · applicability & verification plan

What we do not know yet about CodexMonitor

  • iOS support is work in progress; terminal and dictation are unavailable there.
  • CLI sessions do not stream live until resumed.

Applicability

Repository documentation at dd61b9abd37de5ded86e82b9fe8a83fd49d46fa5; Unpinned public documentation and/or pinned README sections reviewed 2026-09-13; per-claim source scope applies.

Documentation review only

Excludes: Runtime behavior; Performance comparisons; Security guarantees

Recheck when

  • Documentation changes
  • New product release
  • License change

Needs recheck · due

Verification plan

  1. Pin a product release or repository commit and record the host environment.
  2. Create a disposable repository with a passing baseline test.
  3. Run a bounded change with two agents where supported; record permissions and logs.
  4. Record a failed attempt, actual outcome, and artifacts without secrets.
  5. Have another person reproduce the result before labeling it reproduced.

Provenance & review

Testedknowhow initial research

Commissioned by a Nimbalyst maintainer; not an independent comparison.

Revision 2026-09-13.4 · unreviewed · checked

0 attestations · Content: Unlicensed prototype; redistribution decision pending

cmux →

Needs recheck · 2026-10-130 attestations · rev 2026-09-13.4 · unreviewed
Restart recoveryPartial
Explore Restart recovery →

Restores layout and resumes supported agents using saved native IDs.

Scope: Hooks required for agent resume. Arbitrary live processes are not checkpointed.

2 open gaps · applicability & verification plan

What we do not know yet about cmux

  • Ordinary process memory is not checkpointed during session restore.
  • README calls the project GPL-licensed; exact license terms need separate review.

Applicability

Repository documentation at 22ae1c56fd83103e2d9c80ec3341a9a08557e6e1; Unpinned public documentation and/or pinned README sections reviewed 2026-09-13; per-claim source scope applies.

Documentation review only

Excludes: Runtime behavior; Performance comparisons; Security guarantees

Recheck when

  • Documentation changes
  • New product release
  • License change

Needs recheck · due

Verification plan

  1. Pin a product release or repository commit and record the host environment.
  2. Create a disposable repository with a passing baseline test.
  3. Run a bounded change with two agents where supported; record permissions and logs.
  4. Record a failed attempt, actual outcome, and artifacts without secrets.
  5. Have another person reproduce the result before labeling it reproduced.

Provenance & review

Testedknowhow initial research

Commissioned by a Nimbalyst maintainer; not an independent comparison.

Revision 2026-09-13.4 · unreviewed · checked

0 attestations · Content: Unlicensed prototype; redistribution decision pending

Sources · 1

[1]cmux · readme ↗Pinned12 / 13 claims
repository · locator & rights

22ae1c56fd83103e2d9c80ec3341a9a08557e6e1

README feature sections; license and limitations at this pinned revision.

Authored summary and link only; upstream rights retained.

Cline →

Needs recheck · 2026-10-130 attestations · rev 2026-09-13.4 · unreviewed
Restart recoveryPartial
Explore Restart recovery →

Scheduled agents persist across restarts.

Scope: Schedules only; no claim that an arbitrary in-flight tool call survives.

2 open gaps · applicability & verification plan

What we do not know yet about Cline

  • Capabilities span several clients and the SDK; parity needs testing.
  • Repository notes JetBrains plugin code is not open-sourced.

Applicability

Repository documentation at c1c0b55ca03a4cd1d1eff2f64449445f778be4c4; Unpinned public documentation and/or pinned README sections reviewed 2026-09-13; per-claim source scope applies.

Documentation review only

Excludes: Runtime behavior; Performance comparisons; Security guarantees

Recheck when

  • Documentation changes
  • New product release
  • License change

Needs recheck · due

Verification plan

  1. Pin a product release or repository commit and record the host environment.
  2. Create a disposable repository with a passing baseline test.
  3. Run a bounded change with two agents where supported; record permissions and logs.
  4. Record a failed attempt, actual outcome, and artifacts without secrets.
  5. Have another person reproduce the result before labeling it reproduced.

Provenance & review

Testedknowhow initial research

Commissioned by a Nimbalyst maintainer; not an independent comparison.

Revision 2026-09-13.4 · unreviewed · checked

0 attestations · Content: Unlicensed prototype; redistribution decision pending

Sources · 1

[1]Cline · readme ↗Pinned19 / 19 claims
repository · locator & rights

c1c0b55ca03a4cd1d1eff2f64449445f778be4c4

README feature sections; license and limitations at this pinned revision.

Authored summary and link only; upstream rights retained.

Roo Code →

Needs recheck · 2026-10-130 attestations · rev 2026-09-13.4 · unreviewed

Unknown — no evidence recorded

2 open gaps · applicability & verification plan

What we do not know yet about Roo Code

  • Repository is archived; README states the extension was shut down on May 15.
  • Treat feature claims as historical; investigate a maintained alternative before adoption.

Applicability

Repository documentation at b867ec9145750d0ae1ff7f02d35406e9bf2a0b16

Documentation review only

Excludes: Runtime behavior; Performance comparisons; Security guarantees

Recheck when

  • Documentation changes
  • New product release
  • License change

Needs recheck · due

Verification plan

  1. Pin a product release or repository commit and record the host environment.
  2. Create a disposable repository with a passing baseline test.
  3. Run a bounded change with two agents where supported; record permissions and logs.
  4. Record a failed attempt, actual outcome, and artifacts without secrets.
  5. Have another person reproduce the result before labeling it reproduced.

Provenance & review

Testedknowhow initial research

Commissioned by a Nimbalyst maintainer; not an independent comparison.

Revision 2026-09-13.4 · unreviewed · checked

0 attestations · Content: Unlicensed prototype; redistribution decision pending

Kilo Code →

Needs recheck · 2026-10-130 attestations · rev 2026-09-13.4 · unreviewed

Unknown — no evidence recorded

2 open gaps · applicability & verification plan

What we do not know yet about Kilo Code

  • Cloud Agent and code-review services are separate hosted surfaces.
  • CLI autonomous mode disables confirmations; it is not a sandbox.

Applicability

Repository documentation at c36e22634860e06e0aa63234fae37bbd83d3b182

Documentation review only

Excludes: Runtime behavior; Performance comparisons; Security guarantees

Recheck when

  • Documentation changes
  • New product release
  • License change

Needs recheck · due

Verification plan

  1. Pin a product release or repository commit and record the host environment.
  2. Create a disposable repository with a passing baseline test.
  3. Run a bounded change with two agents where supported; record permissions and logs.
  4. Record a failed attempt, actual outcome, and artifacts without secrets.
  5. Have another person reproduce the result before labeling it reproduced.

Provenance & review

Testedknowhow initial research

Commissioned by a Nimbalyst maintainer; not an independent comparison.

Revision 2026-09-13.4 · unreviewed · checked

0 attestations · Content: Unlicensed prototype; redistribution decision pending

OpenCode →

Needs recheck · 2026-10-130 attestations · rev 2026-09-13.4 · unreviewed

Unknown — no evidence recorded

2 open gaps · applicability & verification plan

What we do not know yet about OpenCode

  • Desktop app is labeled beta in the cited README.
  • A coding runtime is adjacent to, not equivalent to, a multi-runtime management platform.

Applicability

Repository documentation at df23b7f9488a38e6f8064a0739d4f8cde86d7cfb; Unpinned official documentation checked 2026-09-13; Unpinned public documentation and/or pinned README sections reviewed 2026-09-13; per-claim source scope applies.

Documentation review only

Excludes: Runtime behavior; Performance comparisons; Security guarantees

Recheck when

  • Documentation changes
  • New product release
  • License change

Needs recheck · due

Verification plan

  1. Pin a product release or repository commit and record the host environment.
  2. Create a disposable repository with a passing baseline test.
  3. Run a bounded change with two agents where supported; record permissions and logs.
  4. Record a failed attempt, actual outcome, and artifacts without secrets.
  5. Have another person reproduce the result before labeling it reproduced.

Provenance & review

Testedknowhow initial research

Commissioned by a Nimbalyst maintainer; not an independent comparison.

Revision 2026-09-13.4 · unreviewed · checked

0 attestations · Content: Unlicensed prototype; redistribution decision pending

Aider →

Needs recheck · 2026-10-130 attestations · rev 2026-09-13.4 · unreviewed

Unknown — no evidence recorded

2 open gaps · applicability & verification plan

What we do not know yet about Aider

  • Aider is a coding runtime rather than a team session manager.
  • Web-page context ingestion is not counted as browser automation.

Applicability

Repository documentation at 5dc9490bb35f9729ef2c95d00a19ccd30c26339c; Unpinned public documentation and/or pinned README sections reviewed 2026-09-13; per-claim source scope applies.

Documentation review only

Excludes: Runtime behavior; Performance comparisons; Security guarantees

Recheck when

  • Documentation changes
  • New product release
  • License change

Needs recheck · due

Verification plan

  1. Pin a product release or repository commit and record the host environment.
  2. Create a disposable repository with a passing baseline test.
  3. Run a bounded change with two agents where supported; record permissions and logs.
  4. Record a failed attempt, actual outcome, and artifacts without secrets.
  5. Have another person reproduce the result before labeling it reproduced.

Provenance & review

Testedknowhow initial research

Commissioned by a Nimbalyst maintainer; not an independent comparison.

Revision 2026-09-13.4 · unreviewed · checked

0 attestations · Content: Unlicensed prototype; redistribution decision pending

Warp →

Needs recheck · 2026-10-130 attestations · rev 2026-09-13.4 · unreviewed

Unknown — no evidence recorded

2 open gaps · applicability & verification plan

What we do not know yet about Warp

  • Factory capabilities and access depend on the configured offering.
  • Vendor outcome and cost claims have not been reproduced.

Applicability

Unpinned public documentation as of 2026-09-13

Documentation review only

Excludes: Runtime behavior; Performance comparisons; Security guarantees

Recheck when

  • Documentation changes
  • New product release
  • License change

Needs recheck · due

Verification plan

  1. Pin a product release or repository commit and record the host environment.
  2. Create a disposable repository with a passing baseline test.
  3. Run a bounded change with two agents where supported; record permissions and logs.
  4. Record a failed attempt, actual outcome, and artifacts without secrets.
  5. Have another person reproduce the result before labeling it reproduced.

Provenance & review

Testedknowhow initial research

Commissioned by a Nimbalyst maintainer; not an independent comparison.

Revision 2026-09-13.4 · unreviewed · checked

0 attestations · Content: Unlicensed prototype; redistribution decision pending

Zed →

Needs recheck · 2026-10-130 attestations · rev 2026-09-13.4 · unreviewed

Unknown — no evidence recorded

2 open gaps · applicability & verification plan

What we do not know yet about Zed

  • Native, ACP, and terminal agents have different configuration paths.
  • Model access does not imply every external agent supports the same tools.

Applicability

Unpinned public documentation as of 2026-09-13

Documentation review only

Excludes: Runtime behavior; Performance comparisons; Security guarantees

Recheck when

  • Documentation changes
  • New product release
  • License change

Needs recheck · due

Verification plan

  1. Pin a product release or repository commit and record the host environment.
  2. Create a disposable repository with a passing baseline test.
  3. Run a bounded change with two agents where supported; record permissions and logs.
  4. Record a failed attempt, actual outcome, and artifacts without secrets.
  5. Have another person reproduce the result before labeling it reproduced.

Provenance & review

Testedknowhow initial research

Commissioned by a Nimbalyst maintainer; not an independent comparison.

Revision 2026-09-13.4 · unreviewed · checked

0 attestations · Content: Unlicensed prototype; redistribution decision pending

Keep exploring session continuity