‹ Back to apps
rinkata
Keep your AI agents aligned with what you decided to build. Rinkata holds your Goals, specs, tickets, and decisions, and flags when code work drifts from them. It connects over MCP with OAuth, so there are no keys to paste.
What’s inside
1 MCP server · 10 skills- MCP serverrinkataMCP server: rinkata
- Skillrinkata-feedbackVolunteer free-form feedback about using rinkata itself — a moment of friction, a confusing error, a surprising or a genuinely great result — by calling the rinkata_submit_feedback MCP tool. This is product feedback to the rinkata team, not project work: it lands in the team’s triage queue (Hub → Admin → Feedback), where it can be resolved or promoted into the inbox → ticket pipeline. Use this skill when the user says "send feedback", "/rinkata-feedback", "tell the rinkata team …", or when you (the agent) hit something worth reporting while using rinkata. A human or agent volunteers the signal directly.
- Skillrinkata-ideaCapture a raw user idea, classify it, and promote it through the canonical rinkata MCP write path. Use this skill for /idea, /rinkata-idea, "capture this idea", "turn this into a Goal", "turn this into a spec", or when a user asks whether an idea should become a Goal, standalone spec, anchored ticket, or spike. Also use it when a user asks to list, review, or find captured ideas. Load mid-session without a slash when an idea is still on Ideas, the user confirms a destination, or they want a successor of a complete spec from an idea — then call rinkata_promote_idea (do not create_standalone_spec + originIdeaId). Specs can now live on their own: a standalone spec is a valid destination when the idea is already concrete enough to be a contract but does not need a Goal section yet. When an idea has HubImages, list/read returns coverImageId / images[] metadata; call rinkata_read_hub_image to vision-read the cover — do not invent a second ceremony.
- Skillrinkata-import-projectBring an existing project under rinkata without guessing. Inventories current Goals/specs/tickets, identifies source-of-truth gaps, and proposes import work.
- Skillrinkata-opsrinkata is a Hub source-of-truth consistency engine. Hub DB is canonical for Goals, specs, tickets, decisions, reconcile artifacts, inbox notes, handoffs, and completion evidence. Installed MCP tools, the CLI, and the Hub UI are the access paths; repo-local artifact directories are not a rinkata surface. Use this skill whenever the user mentions rinkata projects, tickets, preflight, reconcile, drift, doctor, ticket IDs like TICK-*, Needs you, or Decisions. Also load mid-session to mint a proposed Decision (judgment call / log a decision) then list needs_action — do not wait for a slash command.
- Skillrinkata-orientOne-call orient for any rinkata-backed project. Surfaces project truth — counts, drift, doctor verdict, top blockers, and a suggested next action. Always reads Hub truth through the active rinkata MCP surface. Use this skill at session start for any rinkata project, whenever the user asks "what’s the state of this project?" or "what should I work on?", or any time context-restoration is needed mid-session. Read-only: this skill never mutates project state. For mutations (decide, ticket-start, ticket-complete) use the dedicated skills.
- Skillrinkata-reconcileReview and update out-of-date work between Goals, specs, and tickets. Uses rinkata reconcile tools to propose reviewable plans, hands each plan to a human to apply, and treats stale graphs as blockers. Default path for out-of-date Goal/Spec work is staged propose then human apply (not Hub-only). Not for drift findings (Hub: mismatches).
- Skillrinkata-status-reportProduce a project status report from rinkata truth: counts, drift, blockers, in-flight specs, completion gaps, and suggested next actions. Read-only.
- Skillrinkata-ticket-completeClose a rinkata ticket with optional evidence. Reads advisory preflight, then calls rinkata_complete_ticket so status, completed_at, completion_evidence, and the protected Completion block are written by the tool. Load mid-session without a slash when the work has shipped (rinkata_complete_ticket) or the user wants only to record what we now know (rinkata_knowledge_record; do not complete).
- Skillrinkata-ticket-startThe "I’m picking up TICK-X" entry point for any rinkata-backed project. Runs preflight on the ticket, reports advisory alignment and dependency discrepancies, surfaces the ticket body + spec acceptance criteria into the transcript so the agent has context BEFORE working, then claims the ticket — assigns it to the caller and transitions it todo → in_progress in one call via rinkata_claim (official Start). Just before the Start it binds the host session to the ticket (rinkata_bind_usage_session), so its measured token usage counts from the start of the work; a resumed in_progress ticket gets the bind alone. After a successful Start, always reports server `filledParents` (print kind+id entries, or state that the array is empty — never invents fills). Use this skill when an engineer or AI agent is about to start work on a specific ticket — typically right after /rinkata-orient suggests one. /rinkata-orient answers "what should I work on"; /rinkata-ticket-start claims the chosen ticket and packs its context. Also load mid-session without a slash when the user is picking up, starting, or claiming a ticket (claim vs start: rinkata_claim start:false vs default start:true). Mutating: this skill flips ticket status. It never flips status as a side effect of any other action — only on its own explicit invocation.
- Skillrinkata-trash-restoreRecover or inspect archived rinkata work without silent resurrection. Surfaces archived specs/tickets and requires an explicit restore/re-anchor decision.