Issues & AI remediation

POST   /api/issues                         # user: report a problem; requires instance_id plus media scope; tmdb/tvdb ids.
                                           #   Non-admins may only report against a library they can see
                                           #   (403 in requester vocabulary otherwise); admins may report anywhere
GET    /api/issues                         # reporter inbox: the caller's OWN reports, newest first, requester copy applied
GET    /api/admin/agent-digest             # the agent scoreboard. "Resolved" renders as OUTCOME (real-work resolved + self-cleared) with disjoint attribution lanes (by rules / by agent / by admin; remainder "on their own"); "Close without fix" and dismissals counted but never called resolved; plus actions executed, tokens, and needs-a-human/paused counts that are state NOW, not window facts
GET    /api/admin/agent-approval-rules/candidates  # triples the admin has hand-approved and could automate (grounded; active-ruled triples excluded)
POST   /api/admin/agent-approval-rules     # arm a rule from the catalog; server re-checks the triple was actually hand-approved
                                           #   for movie/tv, foreign_id (+book_format when both formats exist) for book
                                           #   (instance must be Radarr for movie, Sonarr for tv, Chaptarr for book;
                                           #   gated by allow_reporting); returns {issue_id,status}; initial status is
                                           #   observing or recovering
GET    /api/issues/{id}                    # reporter or admin: issue thread (an admin viewing marks it read)
POST   /api/issues/{id}/reply              # reporter or admin: reply (answers agent questions)
POST   /api/issues/{id}/confirm-fixed      # reporter ONLY: close their own report as fixed
GET    /api/admin/issues?status=&closed_limit=  # admin: issue queue (user-reported + auto-detected; each row carries read/unread). Open issues always complete; closed history bounded (default 200, max 1000) with closed_total for what is not shown
POST   /api/admin/issues/{id}/dismiss      # admin
POST   /api/admin/issues/{id}/resolve      # admin: { disposition: resolved|wont_fix, note?: <=8192 bytes }; blank gets attributed default audit text
                                           #   transactional reviewed completion; races return 409
POST   /api/admin/issues/{id}/reopen       # admin ONLY: no body; restore manual review; 409 links existing_issue_id for a duplicate
GET    /api/admin/issues/{id}/activity     # admin: durable action + run history for one issue
GET|PUT /api/admin/remediation-settings    # admin: master switch, auto-dispatch, reporting,
                                           #   mark-resolved-as-read, mode, optional model_override
                                           #   (provider/credential always follow the shared selection),
                                           #   step/turn/time and daily-run budgets,
                                           #   observation_min_minutes (10), observation_quiet_minutes (5),
                                           #   observation_settle_minutes (2)
GET    /api/admin/agent-actions?status=    # admin: awaiting queue, or status=all for history;
                                           #   actions include immutable instance_id + instance name/service;
                                           #   release GUIDs are one-way fingerprints, never raw indexer capabilities
GET    /api/admin/agent-actions/{id}       # admin: reconcile one durable action after a lost response
POST   /api/admin/agent-actions/{id}/approve   # admin: claims and dispatches the stored proposal once;
                                           #   body {override?, remember?}; remember additionally arms a
                                           #   standing auto-approval rule for this (problem, fix, facet)
POST   /api/admin/agent-actions/approve-batch  # admin: approve an explicit list of reviewed proposals;
                                           #   body {ids} (≤100), decided sequentially by the single-
                                           #   approve core; 200 with per-item verdicts (durable action
                                           #   status, or skipped/error with a reason); one conflict
                                           #   never fails the rest
POST   /api/admin/agent-actions/{id}/deny      # admin: denial resumes the investigation
GET    /api/admin/agent-runs/{id}          # admin: full audit trail of one agent run
GET    /api/admin/agent-approval-rules     # admin: standing auto-approval rules with status + counters
POST   /api/admin/agent-approval-rules/{id}/pause    # admin: stop a rule matching (idempotent)
POST   /api/admin/agent-approval-rules/{id}/resume   # admin: re-arm a paused rule (idempotent)
DELETE /api/admin/agent-approval-rules/{id}          # admin: remove a rule; history keeps its attribution

View the maintained source for this page.