chore: release packages

[github-actions]

This PR was opened by the Changesets release GitHub action. When you're ready to do a release, you can merge this and the packages will be published to npm automatically. If you're not ready to do a release yet, that's fine, whenever you add more changesets to main, this PR will be updated.

Releases

@getmunin/backend-core@4.77.0

Minor Changes

• 2d14917: fix(outreach): keep outreach_list_proposals payloads bounded

  outreach_list_proposals returned every column of every matching row, curator evidence included. Evidence is an unbounded JSONB the curator fills with sources, compliance notes and reasoning — around 4,000 characters per proposal in practice, roughly three quarters of a row. Combined with a default limit of 100 and a default of all statuses, a queue of ~16 proposals already produced an 80,000-character result that clients refuse, and 100 rows would have been half a million characters. The failure is silent-ish and total: the MCP Apps panel renders the size error instead of the review UI, and the model gets no data either, so the review pass just stops.

  List rows now carry the draft, the nested contact / campaign / delivery summaries and a boolean hasEvidence, but not evidence itself. The default limit drops from 100 to 25 and the ceiling from 500 to 200.

  The new outreach_get_proposal reads one proposal by id with the full evidence attached, so nothing became unreachable — this exposes the getProposal service method the Slack bridge and GET /v1/outreach/proposals/:id already used. The Inspector panel's Evidence toggle now fetches on click rather than receiving evidence for every card up front.

  GET /v1/outreach/proposals and the inbox queue return the same trimmed rows. Nothing in the dashboard rendered evidence from a list response.

• cfa7b4f: refactor(outreach): first-touch replaces "initial" on the public surface

  outreach_propose_initial_message is now outreach_propose_first_touch, the campaign flag autoDraftInitial is now autoDraftFirstTouch (column auto_draft_initial → auto_draft_first_touch), and the three skills skill://outreach/draft-initial-{email,sms,call} are now skill://outreach/draft-first-touch-{email,sms,call}.

  The three propose tools file the three proposal kinds (initial, reply, followup), but only the first carried a medium in its name — _message was filler to make outreach_propose_initial grammatical, and inaccurate besides: the tool also files the script for an outbound voice call, which is not a message. Its description now says "first-touch outreach draft", the neutral term the input schema already uses (draftSubject / draftBody). The campaign flag had the same defect: autoDraftInitial paired with autoDraftReplies put a bare adjective next to a noun, while the surrounding descriptions had already switched to saying "first-touch".

  The weekly scheduled sweep follows: it is now curator-outreach-first-touch, reads MUNIN_CURATOR_OUTREACH_FIRST_TOUCH_CRON, and enqueues under the dedupe key outreach-first-touch:scheduled.

  Internals that track the stored kind keep initial: the outreach_proposals.kind value and OutreachService.proposeInitial.

  Breaking, with no aliases published:

  • callers that hardcode outreach_propose_initial_message
  • callers that send or read autoDraftInitial on outreach_create_campaign / outreach_update_campaign / outreach_list_campaigns
  • callers that read the old skill://outreach/draft-initial-* URIs via skills_read / resources/read
  • self-hosters who set MUNIN_CURATOR_OUTREACH_INITIAL_CRON — the old name is ignored and the sweep silently reverts to its weekly default, so rename it
  • campaign JSON exported before this release fails outreach_import validation on the renamed field

  Migration 0059_outreach_first_touch_rename renames the column in place, so stored per-campaign values survive, and carries the persisted curator queue across. job_uri, dedupe_key and source_event_type are rewritten on every row — they point at things that were renamed, so history stays queryable and a pending row still dedupes against the next scheduled enqueue. user_prompt is rewritten only for status = 'pending' rows: the scheduler persists its prompt verbatim at enqueue time, so a queued job would otherwise wake up naming a tool and a skill that no longer exist, while a finished job's prompt is the record of what it was actually told. Every step is guarded and safe to re-run.

• d0d3d28: Rename conv_send_voice_sms_channel_test → conv_send_sms_channel_test

  The send path dispatches on capabilities.sendTest, which only Twilio and MessageBird set. Vapi and Threll both declare sendTest: false, so on a voice channel the tool has always answered channel vendor 'vapi' does not support test sends — the voice_sms qualifier 4.76.0 gave it promised a path that does not exist.

  Its sibling conv_test_voice_sms_channel keeps the qualifier: all four vendors implement the credential check. The pair reads asymmetrically now, which is the point — their reach differs.

  The description says SMS-only and points voice at conv_test_voice_sms_channel for credentials and the dashboard's Make a test call for end-to-end. skill://conv/setup-voice-sms-channel names the error a voice channel gets. There is still no MCP tool that places a call: ChannelAdminService.call() stays /v1-only, human-initiated.

Patch Changes

• Updated dependencies [2d14917]
• Updated dependencies [cfa7b4f]
  • @getmunin/inspector-app@4.77.0
  • @getmunin/types@4.77.0
  • @getmunin/db@4.77.0
  • @getmunin/agent-runtime@4.77.0
  • @getmunin/core@4.77.0
  • @getmunin/mcp-toolkit@4.77.0
  • @getmunin/emails@4.77.0

@getmunin/dashboard-pages@4.77.0

Minor Changes

• 2d14917: fix(outreach): keep outreach_list_proposals payloads bounded

  outreach_list_proposals returned every column of every matching row, curator evidence included. Evidence is an unbounded JSONB the curator fills with sources, compliance notes and reasoning — around 4,000 characters per proposal in practice, roughly three quarters of a row. Combined with a default limit of 100 and a default of all statuses, a queue of ~16 proposals already produced an 80,000-character result that clients refuse, and 100 rows would have been half a million characters. The failure is silent-ish and total: the MCP Apps panel renders the size error instead of the review UI, and the model gets no data either, so the review pass just stops.

  List rows now carry the draft, the nested contact / campaign / delivery summaries and a boolean hasEvidence, but not evidence itself. The default limit drops from 100 to 25 and the ceiling from 500 to 200.

  The new outreach_get_proposal reads one proposal by id with the full evidence attached, so nothing became unreachable — this exposes the getProposal service method the Slack bridge and GET /v1/outreach/proposals/:id already used. The Inspector panel's Evidence toggle now fetches on click rather than receiving evidence for every card up front.

  GET /v1/outreach/proposals and the inbox queue return the same trimmed rows. Nothing in the dashboard rendered evidence from a list response.

• cfa7b4f: refactor(outreach): first-touch replaces "initial" on the public surface

  outreach_propose_initial_message is now outreach_propose_first_touch, the campaign flag autoDraftInitial is now autoDraftFirstTouch (column auto_draft_initial → auto_draft_first_touch), and the three skills skill://outreach/draft-initial-{email,sms,call} are now skill://outreach/draft-first-touch-{email,sms,call}.

  The three propose tools file the three proposal kinds (initial, reply, followup), but only the first carried a medium in its name — _message was filler to make outreach_propose_initial grammatical, and inaccurate besides: the tool also files the script for an outbound voice call, which is not a message. Its description now says "first-touch outreach draft", the neutral term the input schema already uses (draftSubject / draftBody). The campaign flag had the same defect: autoDraftInitial paired with autoDraftReplies put a bare adjective next to a noun, while the surrounding descriptions had already switched to saying "first-touch".

  The weekly scheduled sweep follows: it is now curator-outreach-first-touch, reads MUNIN_CURATOR_OUTREACH_FIRST_TOUCH_CRON, and enqueues under the dedupe key outreach-first-touch:scheduled.

  Internals that track the stored kind keep initial: the outreach_proposals.kind value and OutreachService.proposeInitial.

  Breaking, with no aliases published:

  • callers that hardcode outreach_propose_initial_message
  • callers that send or read autoDraftInitial on outreach_create_campaign / outreach_update_campaign / outreach_list_campaigns
  • callers that read the old skill://outreach/draft-initial-* URIs via skills_read / resources/read
  • self-hosters who set MUNIN_CURATOR_OUTREACH_INITIAL_CRON — the old name is ignored and the sweep silently reverts to its weekly default, so rename it
  • campaign JSON exported before this release fails outreach_import validation on the renamed field

  Migration 0059_outreach_first_touch_rename renames the column in place, so stored per-campaign values survive, and carries the persisted curator queue across. job_uri, dedupe_key and source_event_type are rewritten on every row — they point at things that were renamed, so history stays queryable and a pending row still dedupes against the next scheduled enqueue. user_prompt is rewritten only for status = 'pending' rows: the scheduler persists its prompt verbatim at enqueue time, so a queued job would otherwise wake up naming a tool and a skill that no longer exist, while a finished job's prompt is the record of what it was actually told. Every step is guarded and safe to re-run.

Patch Changes

• 2808e5d: Eyebrows no longer echo the heading they sit above. The OAuth consent outcome screens paired "Authorization granted" with the H1 "Access granted." — and in Norwegian both collapsed to the identical string "Tilgang gitt", since tilgang covers both authorization and access. They now name the flow state instead ("Authorization complete" / "Autorisering fullført", "Authorization cancelled" / "Autorisering avbrutt"), leaving the ✓ glyph and the heading to carry the outcome.

  Two inspector panels repeated a noun the same way and now add information instead: the outreach proposals eyebrow reads "Awaiting approval" (was "Pending proposals", above "Outreach proposals"), and the merge panel reads "Possible duplicates" (was "Pending merge proposals", above "Contact merges").

• Updated dependencies [cfa7b4f]

  • @getmunin/types@4.77.0
  • @getmunin/ui@4.77.0

@getmunin/db@4.77.0

Minor Changes

• cfa7b4f: refactor(outreach): first-touch replaces "initial" on the public surface

  outreach_propose_initial_message is now outreach_propose_first_touch, the campaign flag autoDraftInitial is now autoDraftFirstTouch (column auto_draft_initial → auto_draft_first_touch), and the three skills skill://outreach/draft-initial-{email,sms,call} are now skill://outreach/draft-first-touch-{email,sms,call}.

  The three propose tools file the three proposal kinds (initial, reply, followup), but only the first carried a medium in its name — _message was filler to make outreach_propose_initial grammatical, and inaccurate besides: the tool also files the script for an outbound voice call, which is not a message. Its description now says "first-touch outreach draft", the neutral term the input schema already uses (draftSubject / draftBody). The campaign flag had the same defect: autoDraftInitial paired with autoDraftReplies put a bare adjective next to a noun, while the surrounding descriptions had already switched to saying "first-touch".

  The weekly scheduled sweep follows: it is now curator-outreach-first-touch, reads MUNIN_CURATOR_OUTREACH_FIRST_TOUCH_CRON, and enqueues under the dedupe key outreach-first-touch:scheduled.

  Internals that track the stored kind keep initial: the outreach_proposals.kind value and OutreachService.proposeInitial.

  Breaking, with no aliases published:

  • callers that hardcode outreach_propose_initial_message
  • callers that send or read autoDraftInitial on outreach_create_campaign / outreach_update_campaign / outreach_list_campaigns
  • callers that read the old skill://outreach/draft-initial-* URIs via skills_read / resources/read
  • self-hosters who set MUNIN_CURATOR_OUTREACH_INITIAL_CRON — the old name is ignored and the sweep silently reverts to its weekly default, so rename it
  • campaign JSON exported before this release fails outreach_import validation on the renamed field

  Migration 0059_outreach_first_touch_rename renames the column in place, so stored per-campaign values survive, and carries the persisted curator queue across. job_uri, dedupe_key and source_event_type are rewritten on every row — they point at things that were renamed, so history stays queryable and a pending row still dedupes against the next scheduled enqueue. user_prompt is rewritten only for status = 'pending' rows: the scheduler persists its prompt verbatim at enqueue time, so a queued job would otherwise wake up naming a tool and a skill that no longer exist, while a finished job's prompt is the record of what it was actually told. Every step is guarded and safe to re-run.

Patch Changes

• Updated dependencies [cfa7b4f]
  • @getmunin/types@4.77.0

@getmunin/docs-pages@4.77.0

Minor Changes

• cfa7b4f: refactor(outreach): first-touch replaces "initial" on the public surface

  outreach_propose_initial_message is now outreach_propose_first_touch, the campaign flag autoDraftInitial is now autoDraftFirstTouch (column auto_draft_initial → auto_draft_first_touch), and the three skills skill://outreach/draft-initial-{email,sms,call} are now skill://outreach/draft-first-touch-{email,sms,call}.

  The three propose tools file the three proposal kinds (initial, reply, followup), but only the first carried a medium in its name — _message was filler to make outreach_propose_initial grammatical, and inaccurate besides: the tool also files the script for an outbound voice call, which is not a message. Its description now says "first-touch outreach draft", the neutral term the input schema already uses (draftSubject / draftBody). The campaign flag had the same defect: autoDraftInitial paired with autoDraftReplies put a bare adjective next to a noun, while the surrounding descriptions had already switched to saying "first-touch".

  The weekly scheduled sweep follows: it is now curator-outreach-first-touch, reads MUNIN_CURATOR_OUTREACH_FIRST_TOUCH_CRON, and enqueues under the dedupe key outreach-first-touch:scheduled.

  Internals that track the stored kind keep initial: the outreach_proposals.kind value and OutreachService.proposeInitial.

  Breaking, with no aliases published:

  • callers that hardcode outreach_propose_initial_message
  • callers that send or read autoDraftInitial on outreach_create_campaign / outreach_update_campaign / outreach_list_campaigns
  • callers that read the old skill://outreach/draft-initial-* URIs via skills_read / resources/read
  • self-hosters who set MUNIN_CURATOR_OUTREACH_INITIAL_CRON — the old name is ignored and the sweep silently reverts to its weekly default, so rename it
  • campaign JSON exported before this release fails outreach_import validation on the renamed field

  Migration 0059_outreach_first_touch_rename renames the column in place, so stored per-campaign values survive, and carries the persisted curator queue across. job_uri, dedupe_key and source_event_type are rewritten on every row — they point at things that were renamed, so history stays queryable and a pending row still dedupes against the next scheduled enqueue. user_prompt is rewritten only for status = 'pending' rows: the scheduler persists its prompt verbatim at enqueue time, so a queued job would otherwise wake up naming a tool and a skill that no longer exist, while a finished job's prompt is the record of what it was actually told. Every step is guarded and safe to re-run.

Patch Changes

• 688db03: Dim the page behind the open mobile section-nav dropdown in the developer portal, and close it on a tap outside — the panel previously floated over undimmed content, unlike every dialog in the product.
• Updated dependencies [2d14917]
• Updated dependencies [cfa7b4f]
• Updated dependencies [d0d3d28]
  • @getmunin/backend-core@4.77.0

@getmunin/inspector-app@4.77.0

Minor Changes

• 2d14917: fix(outreach): keep outreach_list_proposals payloads bounded

  outreach_list_proposals returned every column of every matching row, curator evidence included. Evidence is an unbounded JSONB the curator fills with sources, compliance notes and reasoning — around 4,000 characters per proposal in practice, roughly three quarters of a row. Combined with a default limit of 100 and a default of all statuses, a queue of ~16 proposals already produced an 80,000-character result that clients refuse, and 100 rows would have been half a million characters. The failure is silent-ish and total: the MCP Apps panel renders the size error instead of the review UI, and the model gets no data either, so the review pass just stops.

  List rows now carry the draft, the nested contact / campaign / delivery summaries and a boolean hasEvidence, but not evidence itself. The default limit drops from 100 to 25 and the ceiling from 500 to 200.

  The new outreach_get_proposal reads one proposal by id with the full evidence attached, so nothing became unreachable — this exposes the getProposal service method the Slack bridge and GET /v1/outreach/proposals/:id already used. The Inspector panel's Evidence toggle now fetches on click rather than receiving evidence for every card up front.

  GET /v1/outreach/proposals and the inbox queue return the same trimmed rows. Nothing in the dashboard rendered evidence from a list response.

Patch Changes

• Updated dependencies [2d14917]
• Updated dependencies [cfa7b4f]
• Updated dependencies [2808e5d]
  • @getmunin/dashboard-pages@4.77.0
  • @getmunin/types@4.77.0
  • @getmunin/ui@4.77.0

@getmunin/types@4.77.0

Minor Changes

• cfa7b4f: refactor(outreach): first-touch replaces "initial" on the public surface

  outreach_propose_initial_message is now outreach_propose_first_touch, the campaign flag autoDraftInitial is now autoDraftFirstTouch (column auto_draft_initial → auto_draft_first_touch), and the three skills skill://outreach/draft-initial-{email,sms,call} are now skill://outreach/draft-first-touch-{email,sms,call}.

  The three propose tools file the three proposal kinds (initial, reply, followup), but only the first carried a medium in its name — _message was filler to make outreach_propose_initial grammatical, and inaccurate besides: the tool also files the script for an outbound voice call, which is not a message. Its description now says "first-touch outreach draft", the neutral term the input schema already uses (draftSubject / draftBody). The campaign flag had the same defect: autoDraftInitial paired with autoDraftReplies put a bare adjective next to a noun, while the surrounding descriptions had already switched to saying "first-touch".

  The weekly scheduled sweep follows: it is now curator-outreach-first-touch, reads MUNIN_CURATOR_OUTREACH_FIRST_TOUCH_CRON, and enqueues under the dedupe key outreach-first-touch:scheduled.

  Internals that track the stored kind keep initial: the outreach_proposals.kind value and OutreachService.proposeInitial.

  Breaking, with no aliases published:

  • callers that hardcode outreach_propose_initial_message
  • callers that send or read autoDraftInitial on outreach_create_campaign / outreach_update_campaign / outreach_list_campaigns
  • callers that read the old skill://outreach/draft-initial-* URIs via skills_read / resources/read
  • self-hosters who set MUNIN_CURATOR_OUTREACH_INITIAL_CRON — the old name is ignored and the sweep silently reverts to its weekly default, so rename it
  • campaign JSON exported before this release fails outreach_import validation on the renamed field

  Migration 0059_outreach_first_touch_rename renames the column in place, so stored per-campaign values survive, and carries the persisted curator queue across. job_uri, dedupe_key and source_event_type are rewritten on every row — they point at things that were renamed, so history stays queryable and a pending row still dedupes against the next scheduled enqueue. user_prompt is rewritten only for status = 'pending' rows: the scheduler persists its prompt verbatim at enqueue time, so a queued job would otherwise wake up naming a tool and a skill that no longer exist, while a finished job's prompt is the record of what it was actually told. Every step is guarded and safe to re-run.

@getmunin/chat-widget@4.77.0

Patch Changes

• Updated dependencies [cfa7b4f]
  • @getmunin/types@4.77.0

@getmunin/agent-host@4.77.0

Patch Changes

• Updated dependencies [2d14917]
• Updated dependencies [cfa7b4f]
• Updated dependencies [d0d3d28]
  • @getmunin/backend-core@4.77.0
  • @getmunin/types@4.77.0
  • @getmunin/db@4.77.0
  • @getmunin/agent-runtime@4.77.0
  • @getmunin/core@4.77.0

@getmunin/agent-runtime@4.77.0

Patch Changes

• Updated dependencies [cfa7b4f]
  • @getmunin/types@4.77.0
  • @getmunin/core@4.77.0

@getmunin/core@4.77.0

Patch Changes

• Updated dependencies [cfa7b4f]
  • @getmunin/types@4.77.0
  • @getmunin/db@4.77.0

@getmunin/mcp-toolkit@4.77.0

Patch Changes

• Updated dependencies [cfa7b4f]
  • @getmunin/types@4.77.0
  • @getmunin/core@4.77.0

@getmunin/analytics-tracker@4.77.0

@getmunin/emails@4.77.0

@getmunin/sdk@4.77.0

@getmunin/ui@4.77.0