CALF Manual
CALF gradient C logo
Controlled Agent Lifecycle Framework

One core. Two ways to work.

A practical guide to CALF 1.8.19 Harness and CALF 1.9.16 Workbench. Connect an AI, run a bounded task, inspect the change, and trust only the evidence the checks actually support.

1.8.x · Harness1.9.x · WorkbenchLocal records · English UI · Unicode text

Workbench 1.9.16 · Start with one task

Choose a workspace, connect your existing AI account, describe the result and what to preserve, inspect supported checks, then submit. The first-task guide shows readiness and the next action. Changes, checks and unfinished attempts stay with the task.

Read Current separately from Recorded. A historical success does not confirm changed files. Open Attention only when a human decision is needed.

Actual screens · First task · Current downloads · Legacy releases

Actual Workbench 1.9.16 first-task guide
Actual signed and notarized macOS release; separate example records. No AI task submitted. / 공증된 실제 macOS 실행 파일과 분리된 예제 기록. AI 작업은 제출하지 않았습니다.
Current UNKNOWN and historical SATISFIED
Synthetic technical fixture in the Windows release. Current UNKNOWN and historical SATISFIED are shown separately. / Windows 배포본의 합성 기술 검사 예시. 현재 UNKNOWN과 과거 SATISFIED를 구분합니다.

Engineering references · current release

Harness 1.8.19, Workbench 1.9.16 and SDK 0.2.11 use one Rust reader. Open Engineering knowledge in Workbench when needed, or use the read-only Repository MCP and SDK examples. The 367-entry starter includes entries explicitly marked as names rather than full definitions.

Import a collection only when you choose to. Its index stays local, retains source hashes and status, and loads bounded excerpts. The private 4,334-document qualification dataset is not included. New tasks retain selected references through resume; references never replace file-bound checks or grant permission.

Engineering guide · Optional Ollama/MLX preview and qualification limits · Current release

Earlier screenshots below illustrate existing controls; they are not captures of the new knowledge panel. Local Qwen7B trials stopped under unchanged resource guards and did not complete a task.

공학 지식 참조 · 최신 배포

하네스 1.8.19·워크벤치 1.9.16·SDK 0.2.11이 하나의 Rust 조회기를 사용합니다. Workbench에서 필요할 때 Engineering knowledge를 열거나 읽기 전용 Repository MCP·SDK 예제를 사용하세요. 기본 참조 367개 중 이름만 있는 항목은 전체 정의와 구분합니다.

선택한 자료만 로컬 인덱스로 가져옵니다. 출처 해시와 자료 상태를 보존하고 필요한 본문만 제한해서 읽습니다. 비공개 검증 문서 4,334건은 배포하지 않습니다. 새 작업은 재개할 때도 선택한 참조를 유지하며, 참조가 파일에 결속된 검사나 실행 승인을 대신하지는 않습니다.

공학 지식 안내 · 선택형 Ollama·MLX 미리보기와 검증 한계 · 최신 배포 안내

아래의 이전 스크린샷은 기존 조작 예시이며 새 지식 패널 화면은 아닙니다. 로컬 Qwen7B 시험은 기존 자원 가드가 중단해 작업 완료까지 통과하지 못했습니다.

Native Repository Intelligence

Read the context you need. Keep the original evidence.

Harness 1.8.19 and Workbench 1.9.16 include Repository Intelligence in Rust. In Workbench, open Repository context, connect a CALF history folder and choose one module. Search older decisions or inspect an exact record without loading the whole journal.

Capture local Git saves bounded HEAD/index/working versions and file-relationship candidates. Select a file and its saved version. This does not run project checks or send source to an AI. Historical PASS and static candidates are not current completion evidence.

Repository is no longer an extension. FFmpeg is the only current optional pack. Earlier packages, records, contracts and settings remain available. Full guide and limits · Native Claude connection.

Earlier screenshots elsewhere in this manual illustrate retained workflows. New native Repository controls received separate runtime and screen checks; see the release verification report.

01 / Editions

Choose the interface you need.

Harness · 1.8.19

A native command-line runtime. Use it with existing Codex / Antigravity hooks, or start independent headless tasks. No Tauri window or WebView is loaded.

Workbench · 1.9.16

The same execution core with a Rust + Tauri desktop interface: Task Flow, Git Graph, Diff, Evidence, worker activity, local sources and small checked edits.

New tasks created through calf task and Workbench use the same job format. Point both to the same CALF_WORK_HOME to see the same jobs. Do not run both workers on separate copies of a live records folder.

Legacy 1.8.4 hook records remain in their existing history.html and installation state. They are preserved; they are not automatically imported into Workbench's task list.
02 / Installation

Extract, keep the folder, then launch.

PlatformPackageRequirements
macOS Apple Siliconmacos-arm64.zipmacOS 14+; native ARM64. Open CALF Workbench.app for the desktop edition.
Windows x64windows-x86_64.zipWindows 10/11 x64; Workbench needs Microsoft WebView2 Runtime. Open calf-terminal.exe.
Linux x64linux-x86_64.tar.gzglibc desktop. Workbench requires GTK 3, WebKitGTK 4.1 and a graphical session; Debian 13 is the native UI test environment. Run ./calf-terminal.

Use the package matching the CPU, not only the OS. Intel Mac, Windows ARM64, Linux ARM64 and Alpine/musl are not included. Rust, Node.js and an external SQLite installation are not needed. The checks for your project still need their own runtimes; supported Python checks need Python 3 and, for pytest, an already installed pytest; Cargo checks need an installed Rust toolchain and cached dependencies.

  1. Download the archive and SHA256SUMS from the matching GitHub or GitLab release.
  2. Verify the archive hash, then extract the entire folder. On Linux use tar -xzf FILE.tar.gz to preserve executable permissions.
  3. Keep the core calf / calf.exe beside the desktop executable. On macOS both are inside the app bundle.
  4. Launch Workbench, or run ./calf --help (.\calf.exe --help in PowerShell).
# macOS
shasum -a 256 CALF-Workbench-1.9.16-macos-arm64.zip
# Linux
sha256sum CALF-Workbench-1.9.16-linux-x86_64.tar.gz
# PowerShell
Get-FileHash .\CALF-Workbench-1.9.16-windows-x86_64.zip -Algorithm SHA256
Workbench 1.9.16 for macOS is signed with Developer ID, notarized by Apple and includes its notarization ticket. Harness 1.8.19 uses an ad-hoc integrity signature. Windows packages have no publisher signature. Follow your OS's normal trust controls and your own policy. Do not disable Gatekeeper, SmartScreen or endpoint protections.

The default records folder is below. An explicit absolute CALF_WORK_HOME overrides it. Existing development installations may also have a runtime.json; an explicit runtime takes precedence and fails closed if its home is invalid.

macOS:   ~/Library/Application Support/CALF/workbench
Windows: %LOCALAPPDATA%\CALF\workbench
Linux:   $XDG_DATA_HOME/calf/workbench
         or ~/.local/share/calf/workbench
03 / First task

Choose the work. Keep track of the result.

You choose the folder, existing AI account and result. CALF connects changes, checks and unfinished attempts so you can review what happened.

  1. Choose Choose workspace and select the project CALF may work in.
  2. Choose Connect AI. A saved connection reports local selection, not live authentication or permission.
  3. In an empty request, choose Use a request outline. Fill in Result I want, Keep unchanged and How to check. Review the completed request, then choose Submit.

The guide folds after a new job appears. Reopen Start a task whenever needed. Check setup opens diagnosis on demand. Existing draft text is preserved.

Your project must already have meaningful unittest, configured pytest, supported single-package Cargo integration tests or node:test checks and their tools. The outline does not create tests or acceptance criteria. Review Diff and Evidence before relying on the result.

The existing host retains its account and approval policy. Direct OpenAI API connection is also available; keys use the OS credential store and provider charges may apply. No new live AI-service benchmark is claimed for this UI update.

Records remain local. A remote AI receives the bounded task snapshot and explicitly attached references. Local records do not make remote inference offline.

Check profiles / Python environment

./calf check-discover /absolute/path/project inspects check selection without starting a task. Python selection is .venv → venv → active environment → bounded PATH probe. An invalid explicit environment stops discovery. Dependencies are not installed automatically; pre-existing failures are recorded rather than waived.

Actual Workbench 1.9.16 · First task / 실제 첫 작업 안내
Actual Workbench 1.9.16 · First-task guide / 첫 작업 안내
Actual Workbench 1.9.16 · Draft before submission / 제출 전 초안
Actual Workbench 1.9.16 · Unsubmitted request outline / 제출 전 요청 초안

Isolated macOS home; no AI task submitted. Actual captures; lossless crops retain the original pixels. Computer-wide metrics are not a CALF benchmark.

CALF Workbench 1.9.2 Task Flow · execution and evidence
Workbench 1.9.2 opens a retained, real Codex invoice example: three baseline failures resolved and six registered tests passed. The original run used Harness 1.8.6; this is a record-compatibility demonstration, not a new AI benchmark.
04 / Headless

Run the same lifecycle without the window.

./calf task home
./calf task connect codex
./calf task run "/absolute/path/project" "Fix the total calculation; preserve the tests."
./calf task list
./calf task status TASK_ID

PowerShell uses .\calf.exe. Supply an executable path after connect codex if discovery fails. Shell quotes protect paths and requests containing spaces or CJK text.

CommandBehavior
task submit WORKSPACE REQUESTQueue a new job and print its ID. Does not start a background service.
task run WORKSPACE REQUESTQueue, process eligible jobs in the same home, then return this job's current result. Exit 0 requires SATISFIED; 2 means incomplete; 1 is a command error.
task workProcess the queue in the foreground. Opening Workbench also starts eligible queued work. Its exit 0 means worker idle, not that every task passed.
task pause / resume / cancel IDRequest control. Resume queues the remaining work; run task work afterward. Cancellation is final.
task recheck IDQueue checks after a supported external edit, within the existing budget and scope.
task preferences "TEXT"Save a working preference for future jobs.

For custom non-Python checks and provider adapters, the advanced calf run WORKSPACE REQUEST PROVIDER.json STATE [RESUME_TASK] interface accepts a bounded provider/check specification. The JSON calf api interface exposes the same task operations as Workbench; it requires an explicit home. These interfaces do not add permissions to the connected host.

Share a custom records folder with Workbench
# macOS / Linux: use an existing physical folder
export CALF_WORK_HOME="/absolute/path/calf-records"
./calf task home
# Launch the Workbench executable from this same environment.
# macOS: "CALF Workbench.app/Contents/MacOS/calf-terminal"

# PowerShell
$env:CALF_WORK_HOME = 'D:\CALF\workbench'
.\calf.exe task home
.\calf-terminal.exe

GUI launchers may not inherit terminal environment variables. Keep an existing explicit runtime configuration when using custom storage. Do not move a running records folder.

05 / Host integration

Keep your existing harness and contracts.

Hook integration is optional. It complements the current Codex / Antigravity environment and does not replace the host's approval or sandbox policy.

./calf install --root "/absolute/path/CALF" \
  --codex-home "/absolute/path/.codex" \
  --antigravity-home "/absolute/path/.gemini"
./calf register "/absolute/path/project" "/absolute/path/contract.json" \
  --config "/absolute/path/CALF/config.json"
./calf status "/absolute/path/project" "example-task"

The installer keeps versioned binaries and backs up settings before adding its hooks. Restart the host; Codex may require its normal /hooks review and trust step. Run the exact checks[].commandLine returned by registration. It binds the installed executable, project, task, criterion and contract hash. Keep old installed binaries because old registered commands still reference them.

Use the included contract.example.json as a template. Set a new task ID, relative file scope, an executable-and-arguments array, and environment keys that affect the check. A successful process exit alone is not acceptance; inspect outcome. If no registered evidence exists, CALF retains UNKNOWN.

Remove the hooks without removing records
./calf uninstall --root "/absolute/path/CALF"

If settings still match CALF's recorded edits, the original backup is restored. Later user changes are retained while owned hook entries are removed. Conflicting ownership or backup changes stop removal. Records, backups and versioned binaries remain. Do not use the historical 1.8.0 --detach-only escape unless you understand that original settings restoration cannot be established.

Antigravity compatibility update

Harness 1.8.19 and Workbench 1.9.16 recognize the terminal header in Antigravity’s long-output preview. CALF checks the session, step and transcript location before accepting it. It does not search command output for success text. Earlier evidence and settings stay in place.

For Claude CLI or the desktop app’s local Code mode, use the included hosts/ connection kit and its English or Korean guide. The Claude adapter is tested on macOS.

06 / Workbench

Read activity separately from completion.

AreaWhat it shows
Task FlowRecorded EDIT, CHECK, REPAIR and review stages with small timestamps. Select a node to inspect its evidence.
Git GraphLocal commits, HEAD / branch labels and working-tree status. Read-only observation; no fetch, push, checkout or execution permission is granted.
Worker terminalsRecorded or current operation, process ID, exit information and check evidence. A node can remain in the view after its process has stopped.
System monitorCPU, GPU, memory, swap, disk and network when available. Missing sensors remain unavailable. Whole-system memory is not CALF's own footprint.
JournalAppend-only execution events and recovery evidence. Earlier failure records stay visible after a successful retry.

Colors are paired with text. Green indicates supported completion, amber indicates waiting / incomplete work, and red marks failure or cancellation where applicable. Git graph colors describe repository state, not test success.

CALF Workbench 1.9.2 Task Flow · execution and evidence
Workbench 1.9.2 opens a retained, real Codex invoice example: three baseline failures resolved and six registered tests passed. The original run used Harness 1.8.6; this is a record-compatibility demonstration, not a new AI benchmark.
07 / Diff and evidence

Which exact change did the check verify?

Diff offers Task, Attempt, Recovery and Working Tree views, with Unified and Before & After presentation. Select a file to see the available content, hashes and evidence. Snapshots, text and output have bounds; content that was not retained cannot be reconstructed from a hash.

A CHECK receipt alone is not sufficient. CALF compares the task / attempt, snapshot and current file version where the recorded evidence allows it. A changed file makes earlier success stale. Read the binding message: if exact file-version binding is not established, the viewer must not present that receipt as proof for the displayed change.

PASS / SATISFIED is scoped to the registered requirements, check inputs and supported independent review. It is not proof of every possible requirement or absence of bugs. A review call can use the same provider; it is a separate review invocation, not an independently trusted human or different model.
CALF Workbench 1.9.2 Diff · the exact checked change
Before / After shows the recorded change and its matching CHECK receipt. Light Edit and external-editor controls remain separate from validation.
08 / Small corrections

Light Edit or your own editor.

  1. Select a changed file in Diff.
  2. For a small correction, open Light Edit. Review the text, apply the patch, then wait for revalidation.
  3. For larger changes, select an editor and choose Open in Editor. The detected list is platform-dependent; Add editor lets you select an application or executable manually.
  4. After an external edit, return to CALF and choose Recheck. Old PASS evidence cannot be reused for a different file version.

Light Edit supports files up to 16 KiB and up to 40 changed lines. It checks the expected current hash, refuses protected test/settings files or files outside the editable scope, and uses the existing attempt budget. It is unavailable while the worker is active. Manual editor entries are local; up to eight can be stored. Xcode, VS Code, Cursor, Zed and other supported launchers appear only when detected or manually added.

09 / Recovery

Resume from evidence, not from a success guess.

StateNext step
WAIT / resourcesLeave the worker running or reopen Workbench. Eligible work resumes when resources permit. The fader limits work; it does not drop checks.
RETRY / response incompleteInspect the failure and connection, then Resume. Earlier attempts and consumed budget remain.
PAUSEResume when ready. Pause applies between owned operations, not necessarily at the instant of a click.
CHECK / evidence staleReview the current file version and Recheck. Do not force old evidence to PASS.
CHECK / needs reconciliationAn operation's effects are uncertain. Inspect files and evidence; ordinary Resume is refused until the supported reconciliation is established.
FAIL / budget exhaustedInvestigate the retained attempts. Reuse can create a new task with a fresh request and checks; it does not reset or rewrite the failed task.
STOP / cancelledThe task is final. Start or Reuse a new task if work is needed.

Reuse loads a recipe from a prior task. It does not copy the prior success verdict or grant new authority. A new task is independently checked. Closing Workbench is not a reliable cancellation command: an owned worker can continue. Use Cancel and inspect its state when you need to stop the work.

Legacy index recovery: after backing up records, use calf recover-index WORKSPACE. New job indexes are rebuilt from validated HTML segments when supported. If the authoritative HTML chain is damaged or missing, recovery stops; it must not fabricate the missing events.

10 / Local context

Read locally. Attach deliberately.

  1. Open Local sources and choose Add folder.
  2. Browse or filter names, then preview a text file.
  3. Use Attach to next task only when you want that reference included in the next AI request.
  4. Disconnect a source to remove its CALF connection. Earlier task evidence remains.

Up to four references, within 8 KiB including source details, can be attached. Links, hidden files and credential-related paths are excluded by the bounded reader. Changed attachments must be selected again. This is explicit context selection, not a filesystem sandbox for every tool your existing AI host already has permission to use.

The native calf local-mcp "/absolute/path/folder" command exposes the bounded local reader over stdio MCP for compatible clients. Keep the selected root narrow. It is not a network server, an automatic connection to arbitrary MCP plugins, or permission to send all local documents to an AI.

CALF sources — 실제 Workbench 화면
Historical 1.9.0 UI reference (same workflow layout). A chosen local text file is previewed with Korean, Chinese and Japanese text. The screen explicitly says it has not been sent to AI.
11 / Personalization

Preferences are guidance, not bypasses.

Use My preferences for language, response style and working conventions. New jobs snapshot these preferences into their records. Changing a preference does not rewrite previous jobs or relax a contract. The interface defaults to English; requests, filenames, source text and preferences support Unicode including Korean, Chinese and Japanese, subject to installed fonts.

CALF uses one worker per records home and bounded local data. It does not require local LLMs, an always-running vector database or a multi-agent fleet. Diff and source previews are loaded within limits when needed. Keep sufficient disk space for your retained evidence.

12 / Upgrade safely

Preserve the data; replace only the program.

  1. Let owned operations finish, or request Pause and wait for the owned process to stop. Close the old Workbench window before replacing program files.
  2. Back up the existing installation configuration, history.html, contracts, settings and complete Workbench records home. Keep related journal segments and indexes together; the HTML is authoritative.
  3. Extract the new package into a new folder. Keep the old package until the new one is verified.
  4. For an existing 1.8.4 hook installation, run the 1.8.19 installer with the same installation root and host homes. The installer preserves versioned binaries, bindings and original setting backups. Review hooks through the host's normal mechanism if prompted.
  5. For Workbench, keep the same records home. If the old development build used an explicit runtime.json, retain that private configuration beside the new desktop/core binaries, or launch both interfaces with the same CALF_WORK_HOME. Do not put your runtime configuration in a public package.
  6. Inspect previous records and a newly executed check. Files changing after verification must become stale, not stay PASS.

Earlier shared-core improvements retained: bounded streaming file hashing, owned child shutdown handling, automatic task execution/resume, bounded local journal and preferences, version-bound change/check evidence, and the shared headless task interface. The desktop views remain in 1.9.x. The legacy hook / contract format is retained.

Update-safe Resume: new queued jobs retain their originating runtime by hash. Keep the complete records home, including runtimes. A new launcher routes that job to its retained runtime; it does not rewrite checks, budgets or old evidence. Unverifiable old jobs remain unresolved. Advanced direct calf run states need their original compatible runtime. SDK 0.2.11 can discover and carry the same native profiles.

Rollback: close the new app and retain its data; launch the previous binary against a backup of the appropriate record format after checking compatibility. Do not restore old host settings over later user edits. Never delete versioned binaries still referenced by registered check commands.

Runtime Work · shared contracts and artifacts

Use SDK 0.2.11 to register a Python, Node or loopback HTTP adapter and construct a calf.work.v1 contract. See SDK_WORK_PROTOCOL.md and the runnable examples in the SDK archive. Registration pins declared executables and adapter files; it is not an OS sandbox.

  1. Keep the same CALF_WORK_HOME for Harness, SDK, Local MCP and Workbench.
  2. Open Runtime Work → Import contract…. Import freezes the contract and input snapshots; it does not start execution.
  3. Select the work and use Run / Resume. CALF records the real process, attempt, output manifest and required checks. If an HTTP response is lost, CALF queries the existing request ID before deciding what to do next.
  4. Inspect artifacts, timestamps, checks and Evidence. Recheck validates the current candidate. If the file digest has changed, the result is UNKNOWN.
  5. Use Adopt checked version only for a matching validated candidate, or use a contract that explicitly authorizes automatic adoption. A new attempt consumes the original budget and preserves earlier candidates.

Cancel asks CALF to stop and clean up the work within its configured limits. Until cleanup is confirmed, the work stays UNKNOWN and keeps its execution slot. Reopening the panel reads records and does not execute a task. Lists and evidence are bounded and selected work is loaded on demand.

CALF Workbench 1.9.2 Runtime Work · file artifacts across runtimes
A Python adapter produced a CJK document in the retained 1.9.2 example. A separate check verified the exact UTF-8 content. The adoption record refers to the same file manifest. The same adopted record was verified after reopening Workbench.

Diagnostics, updates and Check Profiles

Open Check setup to inspect the selected provider, executable, workspace and supported check tools. CALF does not install dependencies. Updates verifies a selected release archive and stages a separate copy; keep the existing records home and retained runtimes. Existing queued jobs resume through their recorded original runtime, not an assumed compatible new executable.

JavaScript/TypeScript checks use a bounded node:test profile. A suitable Node runtime is required; TypeScript support depends on the selected runtime. npm scripts and dependency installation are not executed. Python unittest/pytest and bounded single-package Cargo integration checks remain supported.

Source-reader MCP and work MCP are separate. Work MCP runs over stdio in the foreground. Its tools are read-only by default; enable control explicitly when needed. It does not start a new resident service. The current universal protocol covers file artifacts, not arbitrary side effects, remote transactions, monetary accounting or aesthetic quality.

Saved results and current status

Choose a workspace and open Task status. The saved completion result and the current observer verdict appear side by side. Refresh status reads evidence; it does not run a check or call an AI.

A check can have recorded PASS and matching files while the current verdict is UNKNOWN because its execution context differs. For example, LC_ALL: Not set → C.UTF-8 identifies the difference directly. CALF does not change the environment to make it pass.

Managed task cards also have an expandable recorded/current summary. In the SDK, use registered_tasks(workspace) and registered_status(workspace, task). Only the top-level current outcome establishes completion.

Older receipts without bound field details are labelled unavailable. Later observations require an exact full-context digest match and are labelled as later observations; they do not rerun or approve the check. Arbitrary environment values are redacted. Existing CHECKs, contracts and settings are kept.

Actual Workbench status view showing saved and current completion separately
Actual local status-view capture: this observer matched the original context, so both verdicts are SATISFIED. A separate controlled observer with a different LC_ALL displayed UNKNOWN and the exact difference. Capture from the 1.9.4 development view; 1.9.7 carries the same status interface.

Claude Code connection

The lifecycle adapter and versioned setup/update/recovery are now part of the Rust core. Follow Host connections to prepare a native connection beside the previous version, inspect the current settings hash and apply only the reviewed hooks. No Python runtime is needed for this path. Permissions, MCP entries and unrelated hooks are retained.

Claude keeps its own approval flow. Registered checks require real owned-process receipts and matching file versions. A rejected or resolved ordinary tool call cannot create a passing check. New native protocol tests are separate from historical actual desktop/AI sessions; see VERIFICATION.json for the exact scope.

Connect reviewed local MCP tools

Use this extension when an existing work contract needs a local tool. Normal first-task setup is unchanged. Select a pinned server and exact policy; tool descriptions and annotations are data, not permission.

calf_tool_prepare or tool-prepare returns an intent and a review command. It does not execute the requested tool. Review the task, policy hash, arguments, targets, checks and expiry, then use the existing host permission flow. A ticket is not independent proof of a human approval.

Creation uses a new candidate path and runs registered independent checks. The original policy may enable local recovery from the existing terminal receipt and matching files. Recovery never resubmits an uncertain call. Delete, overwrite, external transmission and remote-job recovery remain outside this profile.

MCP contract guide · SDK preparation example · OS-specific evidence

The first-task images in this manual were captured from Workbench 1.9.16; the MCP interface is qualified separately with real local process tests.

Keep the task in focus

Attention lists decisions that need your attention. Open a task to inspect its result; viewing never resumes or approves work. Result summary shows changed files, recorded checks and the current verdict.

Shape request… helps describe the result, what to preserve and how to check it. Add guidance to the current draft, review it and submit through the existing workflow.

Use Find task… or Cmd/Ctrl + K to search the loaded recent tasks, including Korean, Chinese and Japanese. Use arrows to select, Enter to inspect and Esc to close. Draft text and IME composition stay intact.

Light | Dark / Light | Dark changes only the center. Below 1,200 pixels, monitors start folded. Use Show monitors, Close or Cmd/Ctrl + Shift + M to toggle them. Folding pauses system sampling while task execution continues.

Detailed guide · Verification scope. Earlier screenshots retain their original version captions.

13 / Troubleshooting

Know what this release has and has not proven.

SymptomCheck
No automatic checksUse unittest, explicitly configured pytest, or a single Cargo package with integration tests. Install missing tools yourself; unsupported layouts require an explicit advanced check specification.
Codex unavailable / login requiredUse the existing Codex login and select its executable. Do not paste credentials into requests, logs or issues.
App will not openConfirm OS/CPU, extract the whole package, check OS trust prompts and required WebView/GTK libraries. On macOS close any old instance and launch the exact new bundle.
Missing old jobsCompare the records home. Legacy hook history is not the Workbench job list. An invalid explicit home does not silently fall back.
Memory pressure / provider failureRead the specific failure. High whole-system memory does not prove the provider failed from memory pressure.
Editor missingRefresh the list or Add editor manually. Select an installed application/executable, not a document.
UNKNOWN after a changeInspect scope, file hashes, check receipt and cleanup. Resolve the cause; do not delete failure evidence.

The release notes record the actual binary hashes, fresh tests and reused historical evidence. A synthetic provider test proves control flow, not AI service quality. No new long-duration memory qualification is claimed for these packages. CALF is not a security boundary against an attacker running as your OS user and cannot guarantee absolute correctness for arbitrary tasks.

When reporting an issue, include the edition/version, OS/CPU, a minimal reproduction and redacted error text. Do not upload API keys, login state, private runtime files, whole journals or proprietary source without reviewing their contents. Third-party license text is included in THIRD-PARTY-NOTICES.txt; this binary distribution does not publish CALF core source.

Recorded files, quarantine and restore

Open Artifacts to see recorded paths, task references and retention reasons. Select workspace records or a task, then refresh. Reading a plan never moves files.

  1. Choose Preview index quarantine for a verified rebuildable index, or preview an eligible managed file in the file list.
  2. Review its exact path, size and SHA-256, then choose Confirm quarantine.
  3. In Retained moves, choose Review restore and confirm separately. A newer file is never overwritten.

Core-created disposable text has a creation receipt, task references and a retention deadline. A completed task can review release of its own reference. All references must be released and retention must expire before quarantine is eligible. Source, tests, existing logs, candidates, records, contracts, settings and recovery evidence stay protected.

Quarantine retains every byte and does not free disk space. There is no background scan, automatic cleanup or permanent deletion. The optional managed-text API is limited to 16 KiB per file and 128 files per journal. See the full guide. Some screenshots illustrate earlier releases; this section describes the current controls.

IDE에서 CALF 열기 / Open CALF from your IDE

선택형 CALF Connect의 설치 안내와 실제 화면 매뉴얼을 따르세요. 네 IDE × 세 OS의 메뉴·설정·읽기 전용 조회·Workbench 실행을 확인했습니다. 연결은 작업을 자동 제출하지 않습니다.

Actual Workbench 1.9.16 release windows with separate example records. The 12 IDE menu/launch flows were qualified on the unchanged connector runtime; these refreshed screens illustrate the new Workbench UI. No AI task was submitted.

macos

Actual macos Workbench qualification screen

windows

Actual windows Workbench qualification screen

linux

Actual linux Workbench qualification screen