All notable changes to this project will be documented in this file.
[0.3.5] - 2026-11-07
📚 Documentation
Name the jeeves-watcher-openclaw 0.18.0 tools (watcher_list_issues, watcher_exclude_vcs_path, watcher_list_vcs_history) in the harvest template and Google Drive guides
[0.3.4] - 2026-11-06
🐛 Bug Fixes
(admin) Session refresh stops at an in-place reset
⚙️ Miscellaneous Tasks
Release @karmaniverous/jeeves-scripts-core v0.3.4
[0.3.3] - 2026-11-06
🚀 Features
Export the helpers instances copy; add timeoutMs to gh()
⚙️ Miscellaneous Tasks
Pin nvmrc to the engines floor; build core before tests
Release @karmaniverous/jeeves-scripts-core v0.3.3
[0.3.1] - 2026-11-05
🐛 Bug Fixes
(config) Warn, not fail, when config check can't read the gateway Slack accounts
⚙️ Miscellaneous Tasks
Release @karmaniverous/jeeves-scripts-core v0.3.1
[0.3.0] - 2026-11-04
🚀 Features
(slack) Export normalizeSlackTarget from the package root
(runnables) On_failure in manifests and job deltas
(token-metrics) Read OpenClaw agent DB schema 24
(slack) Slack-backfill-history re-reads a channel without moving the poll position
(config) Jeeves-scripts.json may point at the config directory (#57)
(cli) Run and doctor register the instance root instead of forcing JEEVES_SCRIPTS_CONFIG (#57)
(config) Config check and doctor say where the config came from (#57)
(config) Pure helpers for config set/unset (#57)
(cli) Config set and config unset (#57)
(cli) Init --config-dir writes the config outside the repo and points at it (#57)
(library) Library-sync-mirrors mirrors public repo subdirectories into the content tree (checksum rsync, .meta preserved)
(lib) One TASK-template resolver for core templates (Decision 75)
(dispatchers) Daily-briefing templates, values and summary shape (#33)
(config) PunchList and transcripts blocks (Decision 76)
(punch-list) Hoist the punch list domain (#45)
(transcripts) Transcripts-fetch-video and the jeeves-transcripts skill (#54)
Register punch-list and transcripts, root exports and docs
(transcripts) Cache the API response on a dry run so --live spends no second credit
(lib) Dated-report, run-dirs and server-link report helpers
(slack) Slack-refresh-channel-meta (#49)
Export the report-link helpers from the package root
(email) Export respond DKIM check and address parser from the package root
(lib) Carry the gateway's error code and details in GatewayRpcError (#73)
(admin) LastTurnUsage on the schema-23/24 agent-DB reader (#73)
(admin) Token metrics names an agent's main session main (#73)
(config) ContextThreshold, idleMinutes and excludedKeys on the admin-refresh-sessions delta (#73)
(admin) [breaking] Rebuild admin-refresh-sessions on the gateway and the agent DB (#73)
(admin) Re-check busy with sessions.list activeOnly before each live reset (#73)
(admin) A session is busy while a session it spawned has an active run (#73)
(cli) Doctor fails when Node is below the 24.22.2 floor (#75)
(email) Email-respond allowedSenders accepts *@domain and @.domain patterns (#53)
🐛 Bug Fixes
(github) Github-watch creates watch/ before appending its log
(lib) Read openclaw.json as JSON5; fail on unreadable or invalid configs
(github) Verify the acting gh account after every account switch
(github) Sync-repos follows a changed default branch and fails on sync errors
(slack) Never send oldest 0; a fake Slack that orders history like the real one
(github) Drain-collabs hands back untried items on an account mismatch; test the bot-switch checks
(lib) Accept OpenClaw SecretRefs in openclaw.json token fields
(github) Sync-repos reclone puts the old mirror back if the swap fails
(slack) Validate --before and saved backfill progress; test restarts and the poll position
(lib) Keep secretRefSchema internal
(github) Drain-collabs dequeues one item at a time
(github) Sync-repos guards every reclone step and clears leftover copies each sync
(lib) Resolve OpenClaw env references in token fields; any object is a SecretRef
Skill says to switch to the bot first; template pins basic-ftp; helper doc comment back in place
(lib) Env references also resolve from the .env beside openclaw.json and env.vars; name unresolved ones
(github) Drain-collabs queues the mismatched item again
Skill uses a per-command GH_TOKEN for the bot; reclone's opening cleanup only warns
(config) Config paths see own keys only, refuse prototype keys; distinct backups within a second
(cli) Config set/unset keep every backup, restore the env, retry the rename and refuse a concurrent change
(config) The repo's jeeves-scripts.json is legacy however it is named; config check keeps the source when the plugin fails
(cli) Init refers to the schema across drives, keeps the config dir outside the repo, and writes no sample over a legacy config
(template) The sample test writes its own config, so a new instance's npm test passes
People-propose never writes the config file; messages say the instance config
(config) A pointer's config dir must be absolute on this platform
The watcher's map helpers pick up config edits; docs on value quoting
(config) The launcher's instance root reaches the plugin's copy of core
(library) Delete newly excluded files, drop escaping symlinks, refuse a source outside the clone, ignore a stale incoming on a dry run, match site links case-insensitively, refuse .meta in any case
(github) Refuse sources outside the instance root, treat only a 404 as missing, compare by git blob SHA, record last-run after the switch back
(library) Fail a configured run when git or rsync is missing; doctor reports it through a new doctorCheck
(library) Refuse to take over an existing dest the mirror didn't write unless adopt: true; record each live publish
(library) Name the publish record file when it can't be read or isn't valid
(punch-list) Expand {npmGlobal} by the target platform's path rules
(punch-list) Ship both manifests as dry runs; the enabling delta adds --live
(punch-list) Accept {home}/{npmGlobal} input roots and read npm's real global prefix
(punch-list) A harvest dry run leaves no inbox or report behind
(punch-list) Slack.channel takes a conversation id only
(punch-list) No link for a file on another Windows drive
(task-template) Confine template and include names to task-templates/
(briefings) Reference heading and block, run mode, own-directory checks
(slack) Quote proposed config set commands for POSIX shells and PowerShell, run from the instance repo
(slack) A history read cut at its page cap still advances the sweep
(slack) A preview fails on an unusable worker result, opens no DM and deletes no run directory
(slack) Check channel-meta's Slack accounts against the gateway; skip with no loadable config
(lib) Time out the jeeves-server link lookup
(admin) A live reset line no longer says ownership unknown once the active list settled it (#73)
(admin) Session refresh skips OpenClaw's all-zero delivery-mirror copies when reading the last turn (#73)
(admin) The reset notice goes only to Slack-keyed sessions whose origin is Slack (#73)
(email) Email-poll polls every account when one fails, then fails the run naming each (#66)
(email) [breaking] Subdomain senders need DKIM from exactly their own domain (#53)
(email) AllowedSenders takes only exact addresses and *@domain (#53)
(email) Email-poll keeps every failure in the run error (#66)
(cli) A prerelease of the Node floor does not meet it (#75)
💼 Other
Installed git-cliff
PR J: Skill Workshop review job (#70) (#72)
refactor(config): share the report-directory, Slack-account and readable-config rules
Move defaultReportDir (now taking the block's name for its message), isChannelTarget and the gateway Slack account check to config/report-target, and channel-meta's readableConfig to config/readable-config, so the skillReview block (#70) uses the same rules. No behaviour change.
feat(lib): let gatewayRpc set the CLI's wait for the gateway's answer
An optional options.waitMs becomes the openclaw CLI's --timeout (unset: the CLI's 10 s default), so a slow method such as skills.status can wait longer without a shell.
feat(config): add the skillReview block
A typed, optional top-level block for the Skill Workshop review (#70): reportChannel, reportAccount (checked against the gateway's Slack accounts by config check), timezone, serverKey, outputDir (default the report channel's homeDir, checked on the whole config through the shared defaultReportDir) and budgetAlertPercent (1-100, default 90).
feat(admin): hoist the Skill Workshop review as admin-review-skill-proposals
Daily, report-only review of pending Skill Workshop proposals against the installed skill inventory, with a dated report, a numbered Slack summary and the skills catalog budget (OpenClaw 2026.9.6 eb377ac port) plus a quiet-day alert. Applies nothing. Opt-in through the skillReview block; manifest shipped disabled and dry without --live. The 160 description limit is counted in UTF-8 bytes, as OpenClaw enforces it.
Closes #70
test(admin): cover the skill review's unreadable-file, missing-marker, live-line and empty-block branches
docs(admin): name the skill review worker's report file as the code writes it (report.md)
test(admin): kill the skill review's surviving mutants (gateway artifacts, 4-char cap, one long description, live --print-task, kept runs, posting account, punctuation, blank lines)
chore(deps): update vitest, @vitest/coverage-v8, officeparser, rollup and basic-ftp (minor and patch)
refactor(lib): move escapeSlack/decodeSlack to lib/slack-text, shared by slack/channel-meta and admin/skill-review
fix(admin): reject dot-only Workshop proposal ids in the skill review gather
test(admin): a --live skill review run dispatches the worker as live
fix(admin): count the skill review's 160 description limit in chars, a Jeeves catalog-size convention, not an OpenClaw limit
feat(admin): name the proposal id on each line of the skill review post, so a reply's numbers map to ids from the post
docs(admin): state precisely when the catalog port's skill paths differ from OpenClaw's, including the /data container layout
test(admin): kill the review-fix mutants (chars in the snapshot and TASK, revised/added ids in the post's clash set); postId tests in their own file (#72)
[breaking] Require Node >=24.22.2 (#75)
🚜 Refactor
(runnables) Rename src/jobs to src/runnables and defineJob to defineRunnable
(lib) Use retry helpers from @karmaniverous/jeeves
Split core source files over 300 lines (#17)
(email) Use getErrorMessage from @karmaniverous/jeeves in the download
(runnables) Say runnable, not job, in defineRunnable's messages and docs
(slack) Move the channel reader into slack/lib/read-channel
(slack) Backfill-history skips like slack-poll; drop an unreachable branch
(lib) One isEnoent for the three ENOENT checks
Share the acting-login read, move tally counting into the tally module, split library staging out of mirrorTree
(library) Pass the optional mirror deps straight through; build the failure message from the failures
Read a failed gh call's output in one place; compute each failure message once; runMirrors skips through libraryPrerequisite
(library) Drop the platform and deviceOf seams no caller or test sets
(punch-list) State files through core's JSON file helpers
(punch-list) One manifest reader for the runnable tests
Drop two checks the mutation run proved redundant
(task-template) Reuse lib/path-within for the inside-root check
Take the runnable argument parser from @karmaniverous/jeeves (jeeves#151)
(slack) Narrow channel-meta settings by returning the problem
(config) Write the channel-meta report directory rule once; keep the slack schemas pick/omit-safe
(admin) One skip constructor for session refresh's run flow (#73)
(admin) Session refresh closes the agent DBs before the live resets; doctor checks the listed agents (#73)
(email) Put each poll helper's TSDoc on its own helper (#66)
📚 Documentation
(email) Add the adding-a-mailbox procedure to the email guide and skill
(email) Address review on adding a mailbox
(runnables) Runnable vs job wording across guides and skills
(skills) Core package rules, Jira links and ticket edits, channel-status-recap
Clear the TypeDoc warnings
(runnables) Omitting on_failure leaves an existing runner job's target
(runnables) Upgrading to 0.3.0 renames defineJob in the instance
Fix module paths the src/runnables rename left in TSDoc
Jeeves retry helpers reject attempts below 1
(runnables) Sync-jobs updates existing runner jobs from the effective manifest
(slack) Oldest 0, re-reading history and the affected-channel check
(skills) Jeeves-github token requirements, checked against the jobs' calls
List slack-backfill-history; token metrics reads schema 23 or 24
(skills) Jeeves-github token requirements by login, checked against each job's switch
The config lives outside the repo; config set/unset; moving runbook (#57)
(github) Cross-repo file and ruleset sync guide, delta fields and token scopes
(library) Library mirrors guide, config block, runnable id and prerequisites (git, rsync)
Library mirror symlink, exclude, dry-run and site-link rules; github-sync-files source, read, compare and user-owner rules; workflow scope for workflow targets
Github guide covers file and ruleset sync in the indexes, an unreadable acting login and a non-empty target; the publish record's read error; doctorCheck link
(skills) Dispatcher dry-run flag and the CLI form in the new skills
(transcripts) The cache holds the re-serialised response; dry-run credit cost
(slack) Channel meta guide, config and runnables references, jeeves-slack skill
Release audit of guides, README and template against the code
(slack) Channel-meta's account check and no-config skip in the guides
Say the instance config, not jeeves-scripts.json, wherever the config is meant
(slack) Run the channel-meta preview with the repo launcher
(lib) No TSDoc links to names off the root export
(skills) Trim shipped skill descriptions to 160 chars or less
(admin) Session refresh on the gateway and the agent DB (#73)
(admin) [breaking] Admin-refresh-sessions release notes: re-enable with enabled true and --live; main sessions, DMs, child-run busy, Slack-keyed notice (#73)
One @karmaniverous/jeeves import per file after the retry swap
🧪 Testing
(email) Drop headerLines from attachment fixtures
(config) List onFailureSchema in the root export
(meetings) Type the Notion request mock in notion-blocks tests
(meetings) ResolveChromePath tests move to chrome-path.test.ts
(config) Validate on_failure in jobs deltas
(slack) One Slack-ordered history fake in poll.test-helper
(skills) Read skill metadata as JSON5, like OpenClaw; format the skills with Prettier
(github) An invalid github-sync-files delta fails config check and the run (Decision 76)
(github) Share the sync-files world hooks and gh interception in the test helper
(library) Share the mirrorTree fixture and the runnable loader; prove config check with the runnable once; assert the library JSON Schema shape
Unit-test the staging step; cover stdout-only 404s, output-less read failures and empty listings; assert exact values where tests only checked presence
Kill the surviving mutants: payload temp dirs removed, sent/local rsync writes counted, a name shared in one top-level dir left alone, stale incoming cleared, temp dir only on live publishes, clone shallow
(punch-list) Exact assertions instead of bare throws and counts
(punch-list) Kill the surviving mutants
Neutral fixture names in the PR's tests and docs
(punch-list) Split the harvest run tests (shared sandbox, dry-run file)
(punch-list) Cover the live npm prefix read; split the expansion tests
(slack) Batch order takes a channel's oldest wanted field wherever it is listed
(slack) Close channel-meta mutation survivors; generic names in ported fixtures
(slack) Generic ids in the last two ported fixtures
(email) Restore the respond fixture's exact bytes
(slack) Run the channel-meta config test with the runnable's own context type
(library) Give the prototype-parity test room on slow Windows hosts
Add optional pipeline.accounts[].reportOnly. An account's effective flag is emailConfig.reportOnly || accounts[].reportOnly === true, resolved by the new reportOnlyFor(pipeline) (config/report-only.ts) and used by every reader. The email-updates gate (label-actions enqueueEmailUpdates/enqueueLabelActions) drops each action whose own account is report-only; poll, backfill-historical, email-fetch, apply-labels, email-respond and meetings extract/catch-up all take the per-account policy. drain-updates drops (and logs) queued items for an account that is report-only on its own, without calling gog. email-apply-labels skips report-only accounts and refuses only when every target account is report-only.
[51] docs(email): per-account reportOnly in guides, meetings job and jeeves-email skill
⚙️ Miscellaneous Tasks
Release @karmaniverous/jeeves-scripts-core v0.2.2
[0.2.1] - 2026-10-18
💼 Other
[37] fix(slack): save the read position per message and budget each channel's backlog (#37)
slack-poll now processes history oldest first and saves the channel's read position after each top-level message and its replies, so a run killed mid-channel resumes there. A channel stops starting messages after slack.backfillBudgetSeconds (default 60; at least one message always), logs the backfill and hands off, so a backlog can't starve the channels after it. New optional slack.newChannelLookbackDays sets where a channel with no position starts (0: from now; seeded as its position so a quiet channel doesn't slide forward). channel_not_found logs as a SKIP warning. A cursor-store failure still aborts the run.
Closes #37
[37] fix(slack): read history a page at a time within the channel budget (#37)
fetchHistory fetched every page (1.2 s apart) before writing anything, so a long backlog could still outlast the job timeout without saving a position. fetchHistoryPage reads one page just after oldest (Slack anchors to oldest when latest is unset); pollChannel fetches each next page from the position just saved and checks the budget between pages. Addresses Copilot review on #39.
⚙️ Miscellaneous Tasks
Release @karmaniverous/jeeves-scripts-core v0.2.1
[0.2.0] - 2026-10-18
💼 Other
[14-15] fix(dispatchers): --dry-run never starts a worker session, with or without slack (#15)
With slack, --dry-run now reads Slack, builds the TASK and prints a one-line report (job, task length, allowed post targets), then stops. The 'real worker, posts printed' mode is removed. dispatchTaskFile resolves 'previewed' for --dry-run on both paths.
[14-15] test(worker-slack): type the runWorkerSlackJob mock from the module (#15)
(google-drive) Replace inline base64 images with placeholders; add --reconvert (#20)
[14-15] feat(cli): runnables declare their flags; run checks them and prints run --help (#14)
defineJob({ flags, positionals }) declares the accepted flags (lib/args specs, now with descriptions); the job reads them parsed from ctx.flags. run rejects an undeclared option, an unexpected positional or a valued option without a value (exit 1, before the body runs) for a runnable that declares flags, and passes everything through with a deprecation warning for one that declares none (strict for all in a later minor). run --help prints usage, description, flags and the effective manifest without running the body. Every core runnable declares its flags and reads them through ctx.flags (enforced by flags-discipline.test.ts); dispatchers share dispatcherFlags. job.run({ args }) without flags still works: the job parses them itself.
(email) Email-respond assistant-mailbox auto-reply (#22); review and Copilot fixes
⚙️ Miscellaneous Tasks
Release @karmaniverous/jeeves-scripts-core v0.2.0
[0.1.3] - 2026-10-14
🐛 Bug Fixes
(release) GitHub release notes and changelog use the new version tag
⚙️ Miscellaneous Tasks
Release @karmaniverous/jeeves-scripts-core v0.1.3
[0.1.2] - 2026-10-12
🐛 Bug Fixes
(slack) Keep a channel's existing directory when renaming it fails
(slack) Narrow a rename error's code before naming it in the warning
💼 Other
Extended test timeouts
⚙️ Miscellaneous Tasks
Release @karmaniverous/jeeves-scripts-core v0.1.2
[0.1.1] - 2026-10-12
💼 Other
Label the release's own commits with its version in CHANGELOG and release notes
⚙️ Miscellaneous Tasks
Release @karmaniverous/jeeves-scripts-core v0.1.1
[0.1.0] - 2026-10-12
🚀 Features
Add packages/core and packages/template workspaces (Decision 20)
(config) Config schema, silo resolver and config-check CLI ported from jeeves-scripts-template@322054c
(config) [breaking] Drop the deprecated forwardJGS alias; emailConfig.receipt is strict
(config) ImapSecretPath/resolveImapPassword default to paths().imapSecretsDir; port imap-secrets tests (found migrating JGS, Decision 30)
Copy jeeves-scripts-template main@322054c into core wholesale (Decision 32)
(core) Export every built module as @karmaniverous/jeeves-scripts-core/ (instance deep imports during Decision 32)
This reverts commit 9031dd9d701a85aa874684fc38b6bac8deb190bf.
Revert "chore(release): next core release is 0.1.0 (release bumps minor)"
This reverts commit ca9ea201a47fc7d661fd24f252eac011d61a27f9.
(core) Presets authored in TypeScript, shipped compiled (Decision 52)
TypeDoc and release-it configs in TypeScript (Decision 52)
(core) Init names the package after its folder (Decision 53, Q33)
(core) JIRA_BOARD_ID wins over integrations.jira.boardId (Decision 53, Q25)
(core) A manifest's timeout is optional, as in the runner (Decision 54, Q37)
(core) Admin-check-qdrant-health schedule names its time zone (Decision 54, Q39)
(core) Manifest schedules take exactly the runner's forms (Decision 54, Q38)
(core) Config check validates the plugin's local block (Decision 54, Q40)
(core) Config check pins JEEVES_SCRIPTS_CONFIG before loading the plugin (Decision 54, Q40)
(core) Admin-sweep-orphaned-tmp sweeps jobs..extraDirs too (Decision 55, Q43)
🚜 Refactor
Lib, slack, calendar, core, meta, linear, dispatchers and convert pass the strict gates (Decision 32)
X, google-drive, jira and github pass the strict gates (Decision 32)
Email, meetings and admin pass the strict gates; legacy list empty (Decision 32)
(lib) One OpenClaw config reader for gateway and Slack bot tokens
(lib) Split spawn-worker; test spawn retry, gatewayInvoke and the email cache
(linear) Validate the Linear config with Zod; test linearQuery and the paginators
(slack) Split poll.ts into single-purpose modules
(x) Split x-api into client, OAuth, timelines and actions
(meetings,admin) Split extract, doc-fetch and session-refresh
(meetings,email) Split the migrations, fathom-extract and imap/poll
(email) Decompose fetchThreadMetadata
(email) Move the gog account poll out of email/poll
(lib) One OpenClaw state-dir lookup
(slack) Channel workspace lives in the Slack cache
[breaking] Core ships only behaviour every instance needs
(config) [breaking] Remove emailConfig.receipt
(core) Enforce the 300-line limit with max-lines; split the registry and four tests
(core) Settle the root export before v0.1.0 and pin it with a test
(core) GitHub, calendar and email jobs take their collaborators as parameters; no config read at import (review testability refactor, SOLID-7, DRY-9 partial)
(core) Meta, Slack, Google Drive, admin glue and worker-session take their collaborators; tests (review testability refactor, DRY-5, DRY-9)
(core) Meetings take their collaborators; per-file coverage gate; no help subcommand (review testability refactor, COV rows, D57, dead code)
(core) One retry helper, shared walks and fixtures (review DRY-4, DRY-5, DRY-6, DRY-7, DRY-8, DRY-10, DRY-11, WT-2, WT-F1, WT-F2)
📚 Documentation
(config) Export and document every config schema and resolved type (TypeDoc zero warnings)
Google-drive spec and token-metrics runbook move here from the instances (guides/)
Ship README, LICENSE, domain docs, guides and JSON Schema in the core package
Sync core docs and usage text with jeeves-scripts.json and the launcher
Authored docs live in guides/, TypeDoc output stays in docs/
(people) Document the people registry and people propose
(people) Example person is jane-doe
Drop instance-specific names, ids and paths from core
(google-drive) Trim instance spike and verification notes from the design record
(admin) Pending-models add/prune race is self-healing, not lossless
(jobs) Guides/jobs.md (registry, defineJob, manifests, deltas); run resolution and job ids across the guides
Guides and READMEs match the design (plan row 48)
(linear) The Linear config default is {credentialsDir}/linear.json
(core) TSDoc and the email guide name the runnables, not the removed verbs
(guides) Config check registration, doctor states, skills uninstall, override labels, hand-off before npm install