Design docs¶
Verbatim design specs, implementation plans, and standalone runbooks for the tatara platform, copied from the working repo's docs/superpowers/specs/, docs/superpowers/plans/, and top-level docs/ (audit reports, research notes). These are the platform's actual planning artifacts, unedited except where noted in the appendix notes below.
Sorted by date. A design doc and its implementation plan usually share a date and a topic slug (design docs typically end in -design; the plan is the same title without that suffix, or the same date/topic with "Implementation Plan" in the title).
| Date | Title | Scope | Status |
|---|---|---|---|
| 2026-05-24 | Phase 0 - tatara docs repo Implementation Plan | Initialise the tatara docs/architecture repo at /Users/szymonri/Documents/tatara with all baseline files (README, ARCHITECTURE, LICENSE, MEMORY, ROADMAP, CLAUDE.md, .gitignore), commit them on top of the existing... | - |
| 2026-05-24 | Tatara - phases 0 and 1 design | The previous monolithic tatara project (now at ~/Documents/tatara-old) reached v0.3 as one repository containing CRDs, controllers, MCP servers, a bridge, and helm charts. It worked, but the blast radius of any... | Draft for review |
| 2026-05-27 | Argo Workflows CI Migration - Design | Big-bang migration of CI/CD lanes for tatara-cli and tatara-memory from GitHub Actions (and local Makefile targets) to argo workflows triggered via the existing argo-events lane in the homelab. | - |
| 2026-05-27 | Argo Workflows CI Migration - Implementation Plan | Migrate tatara-cli and tatara-memory CI/CD from github actions and local Makefile targets to argo workflows triggered via the existing argo-events lane, with per-step github commit status reporting. | - |
| 2026-05-27 | tatara-cli v0.1.0 MVP - design | Phase 1 (tatara-memory) shipped at v0.1.2 with end-to-end smoke green against real LightRAG v1.4.16. Phase 2 is the platform CLI: | Ready for plan |
| 2026-05-27 | tatara-cli v0.1.0 + tatara-memory v0.1.3 Implementation Plan | Ship the tatara platform CLI: device-flow login, authenticated REST passthrough, and a stdio MCP server exposing all of tatara-memory to Claude Code. Land the prerequisite tatara-memory v0.1.3 (route rename +... | - |
| 2026-05-27 | Tatara Consolidation Phase - Design | 14 tech-debt items across 4 repos plus 1 new repo | - |
| 2026-05-27 | Tatara Consolidation Phase Implementation Plan | Close the 14-item v0.1.x tech-debt backlog across the tatara platform, pivot deploy substrate to a new private tatara-helmfile umbrella repo, and ship tatara-memory v0.2.0 + tatara-cli v0.2.0 carrying the Edge.ID +... | - |
| 2026-06-04 | tatara-claude-code-wrapper - Design | A single-session Claude Code supervisor packaged as a container. One pod runs one persistent interactive claude process driven over a PTY; each turn's result is captured by a custom Stop hook and returned to the... | - |
| 2026-06-04 | tatara-claude-code-wrapper Implementation Plan | A single-session Claude Code supervisor container that drives the | - |
| 2026-06-05 | tatara-chat (phase 8) - design | A REST service serving continuous, per-stream chat rooms so agents in an implementation stream exchange findings and remarks during work, not just at phase or task boundaries. The orchestrator (main agent) spawns a... | approved (brainstorm), pending implementation plan |
| 2026-06-05 | tatara-chat (phase 8) Implementation Plan | Build the tatara-chat Go service - OIDC-gated REST rooms/participants/messages with cursor polling and a 24h TTL sweeper, Postgres-backed, plus a cluster-agnostic helm chart. | - |
| 2026-06-05 | Tatara Code Ingestion - Design | We want to ingest a whole code repository into tatara-memory so agents can ask both structural questions ("who calls Server.Run?", "what does this Terraform resource depend on?", "which templates read... | design approved (forks pinned), pending spec review |
| 2026-06-06 | Tatara as an Agentic-Dev Platform: SOTA Comparison and Improvements | items are mirrored into ROADMAP.md. | notes / improvement backlog input. Not a design spec. Actionable |
| 2026-06-06 | Code-Graph Full Scope - Design (C, M2, Go-fallback, M5) | References:** docs/superpowers/specs/2026-06-05-tatara-code-ingestion-design.md, docs/superpowers/specs/2026-06-06-research-grounded-improvements-design.md (M2/M5),... | design approved (forks pinned); continuous build, checkpoint per merge |
| 2026-06-06 | Code Graph in tatara-memory (sub-project A) Implementation Plan | Add a deterministic code-graph store and query API to tatara-memory: a POST /code-graph:bulk ingest endpoint (file-granular replace) plus GET /code/* traversal endpoints, backed by two new Postgres tables,... | - |
| 2026-06-06 | Code-Graph: Production-Readiness Gap Analysis | what is missing to use the tatara code-graph/memory system against a real (large, multi-repo, legacy) codebase. | assessment. Services deployed; graph empty; pipeline never run on real code. |
| 2026-06-06 | Component C - code-graph MCP tools (tatara-cli) Plan | Add 9 code-graph MCP tools to tatara-cli wrapping tatara-memory's /code/* GET endpoints. | - |
| 2026-06-06 | Go tree-sitter fallback (ingester) Plan | Index non-buildable Go packages via tree-sitter when go/packages reports errors, instead of skipping them. | - |
| 2026-06-06 | M2-A - cross-repo symbols in tatara-memory Plan | Accept cross-repo provides/requires symbol rows on the bulk push, store them per-file-scoped, and serve a /code/cross-repo join endpoint. | - |
| 2026-06-06 | M2-B - cross-repo symbol emission (ingester) Plan | Emit cross-repo provides/requires symbol rows from the Go, Python, and JavaScript analyzers and push them in GraphPush.Symbols. | - |
| 2026-06-06 | M5 - SCIP ingestion (ingester, scip-go first) Plan | Ingest a pre-generated SCIP index (index.scip) into the code graph via a --scip cmd flag, bypassing the source walker. | - |
| 2026-06-06 | Research-Grounded Improvements - Design | for M2/M3/M5) and tatara-chat (C1-C4) Source research: docs/2026-06-06-agentic-dev-sota-improvements.md (SOTA comparison) plus full reads of the primary papers cited per sub-task. | design approved; pending spec review |
| 2026-06-06 | tatara-memory-repo-ingester (sub-project B) - Design | /memories:bulk + /ingest-jobs/{id}), already shipped on tatara-memory main (df1a137). References: the shared contract in docs/superpowers/specs/2026-06-05-tatara-code-ingestion-design.md and the... | design approved; pending spec review |
| 2026-06-06 | tatara-memory-repo-ingester Implementation Plan | Build the Go batch tool that walks a repository, runs best-per-language static analysis (Go, Python, JavaScript, Terraform, Helm, docs), and pushes a deterministic code graph plus enriched semantic chunks to a... | - |
| 2026-06-06 | tatara-operator Design | A Kubernetes operator that orchestrates the tatara platform's unattended agentic-development loop. It manages four CRDs (Project, Repository, Task, Subtask), ingests repositories into tatara-memory,... | approved design, pre-plan. |
| 2026-06-06 | tatara-operator M0 (scaffold) Implementation Plan | Stand up the tatara-operator repo skeleton: kubebuilder/controller-runtime project, the four CRD API types with generated deepcopy, shared internal/{obs,auth,config} packages, a no-reconciler Manager binary,... | - |
| 2026-06-06 | tatara-operator M1 (Project + Repository reconcilers + ingest Job) Implementation Plan | Make tatara-operator ingest repositories into tatara-memory. Deliver two reconcilers and a Job builder: | - |
| 2026-06-06 | tatara-operator M2 (webhook server: push -> main-filtered incremental re-ingest) Implementation Plan | Stand up the operator's webhook server. A POST /operator/webhooks/{project} endpoint on HTTP_ADDR receives GitHub/GitLab events, auto-detects the provider from headers, HMAC-verifies the payload against the... | - |
| 2026-06-06 | tatara-operator M3 (OIDC REST API + tatara-cli MCP tool group) Implementation Plan | Build M3 of tatara-operator: an OIDC-gated REST API (internal/restapi) exposing read/write CRUD over the four CRDs (Project, Repository, Task, Subtask), backed by the controller-runtime client, sharing the... | - |
| 2026-06-06 | tatara-operator M4 (Task reconciler + agent turn loop) Implementation Plan | Build M4 of tatara-operator: the wrapper HTTP Session client (internal/agent/session.go), pure builders for the wrapper agent *corev1.Pod + *corev1.Service (internal/agent/pod.go), the TaskReconciler... | - |
| 2026-06-06 | tatara-operator M5 (SCM write-back + work-item -> Task) Implementation Plan | Complete the three write-side stubs the earlier milestones deliberately left open. (1) Implement scm.GitHub.OpenChange/Comment against the GitHub REST API (POST /repos/{owner}/{repo}/pulls, `POST... | - |
| 2026-06-06 | tatara-operator M6 (chart hardening + deploy wiring) Implementation Plan | Harden and finish the tatara-operator Helm chart (from the M0 skeleton), wire up cluster-agnostic deploy, and stand up the Keycloak + infra-helmfile plumbing so the operator can run in the tatara namespace.... | - |
| 2026-06-06 | tatara-operator shared contracts (pin set) | Authoritative cross-milestone agreements for all tatara-operator plans. The spec (docs/superpowers/specs/2026-06-06-tatara-operator-design.md) is the source of intent; this file pins the exact... | - |
| 2026-06-07 | Enroll tatara (dogfood) Design | Unblock the two image gaps + enroll the tatara platform as the first real operator Project. | approved design, pre-plan. |
| 2026-06-07 | Enroll tatara (dogfood) Implementation Plan | Fix the ingester + wrapper image gaps and enroll the tatara platform as the first real operator Project. | - |
| 2026-06-07 | Per-Project Memory N1 (internal/memory builder package + config) Implementation Plan | Land the pure, unit-tested foundation of per-Project memory provisioning in the deployed tatara-operator repo: | - |
| 2026-06-07 | Per-Project Memory N2 (ProjectReconciler provisioning + status.memory + Ready health + cascade) Implementation Plan | Extend ProjectReconciler so that reconciling a Project provisions its complete per-project memory stack (cnpg Cluster, neo4j StatefulSet+Service, lightrag Deployment+Service+PVC, tatara-memory... | - |
| 2026-06-07 | tatara-operator N3 (per-Project memory endpoint wiring + Ready-gating; remove MEMORY_BASE_URL) Implementation Plan | Make ingest Jobs and agent wrapper pods target the OWNING Project's per-Project memory endpoint (Project.status.memory.endpoint) instead of the removed global MEMORY_BASE_URL, and gate both on the Project's... | - |
| 2026-06-07 | N4 - Retire static tatara-memory + operator chart RBAC/values + image bump + deploy | (helmfile tatara bucket), parent tatara (ROADMAP/MEMORY). | ready to execute. ALL live deploy/uninstall/apply steps are GATED |
| 2026-06-07 | Per-Project Memory Provisioning Design | Make Project creation provision a complete, isolated memory backend (tatara-memory Go service + lightrag + neo4j + cnpg postgres) instead of relying on one shared, statically helm-deployed tatara-memory. Each... | approved design, pre-plan. |
| 2026-06-07 | Per-project memory: shared contracts (pin set) | Authoritative cross-milestone agreements for the per-project memory plans (N1-N4). The spec (docs/superpowers/specs/2026-06-07-per-project-memory-provisioning-design.md) is intent; this pins exact... | - |
| 2026-06-08 | Operator sops-managed secrets Design | Move the operator's cluster-managed secrets from manual kubectl creation to sops-encrypted helmfile values rendered by the tatara-operator chart. | approved design, pre-plan. |
| 2026-06-08 | Operator sops-managed secrets Implementation Plan | Make the tatara-operator chart render its four cluster-managed secrets (tatara-anthropic, tatara-cli-oidc, lightrag-openai, tatara-scm) from sops-encrypted helmfile values, replacing manual `kubectl create... | - |
| 2026-06-09 | Agent-native tatara tools + memory robustness - design | An agent self-audit (issue szymonrychu/tatara-cli#10, PR #11) found that the spawned agent cannot use any tatara MCP tool natively: the pod's .mcp.json is empty and the in-pod tatara CLI has no token (`auth: no... | approved |
| 2026-06-09 | Agent-native tatara tools + memory robustness - Implementation Plan | Let the spawned agent use tatara's own MCP tools natively (headless auth + MCP registration + operator URL), and harden the per-project memory against a postgres blip. | - |
| 2026-06-09 | Namespace-Preserving Clone Layout (tatara-claude-code-wrapper) Implementation Plan | Source specs: | ready to implement |
| 2026-06-09 | Incremental Re-ingest + Phase 0 Capture (tatara-memory-repo-ingester) Implementation Plan | 1. internal/contract/contract.go — mirror the locked Phase 0 wire types (Edge confidence fields, Entity provenance fields, Hyperedge + GraphPush.Hyperedges, new entity-type + relation constants, confidence-tier... | - |
| 2026-06-09 | Phase 0 Contract + Re-ingest Server Surface (tatara-memory) Implementation Plan | 1. Migration wave SQL + migrate.go wiring + MigrationSQL unit tests (codegraph + memory). 2. Contract types: Edge/Entity fields, Hyperedge, GraphPush.Hyperedges, entity-type + relation constants,... | - |
| 2026-06-09 | Incremental Re-ingest (Operator): Full-History Namespace Clone + Per-Repository Cron Schedule -- Implementation Plan | 1. internal/ingest/job.go: drop --depth 1 from cloneCmd (full history so <since> resolves); set clone destination to /workspace/<namespacePath(url)> and point ingester --repo-root at it; add a tiny... | - |
| 2026-06-09 | Cross-repo agent tasks - design | A Task targets one RepositoryRef; the wrapper clones only that repo, and the operator opens one PR on it. The agent does not know the Project's other repos exist, so a cross-repo issue (change spanning... | approved |
| 2026-06-09 | Cross-repo agent tasks - Implementation Plan | Clone every Project repo into the agent workspace and deliver each repo's changes as its own branch + PR, so the agent can handle cross-repo issues. | - |
| 2026-06-09 | Incremental Re-ingest: correct diffs, per-file reconcile, cron schedule | Re-ingestion after the first ingest is broken and incomplete: | design, pending implementation |
| 2026-06-09 | Namespace-preserving clone layout (requirements - spec pending) | tatara-memory-repo-ingester (REPO_ROOT consumption) | requirements captured, design pending (after graphify analysis) |
| 2026-06-09 | Phase 0 contract lock (shared wire types) | Purpose: the exact widened wire types both tatara-memory (consumer) and tatara-memory-repo-ingester (producer, internal/contract/contract.go mirror) must match byte-for-byte. contract_shape_test.go guards drift.... | - |
| 2026-06-09 | Phase 1: structural-parity query tools (no LLM) | Close the read-side graphify gaps that need NO LLM and NO analytics job - they run on the AST data + Phase-0 columns already in prod. Gives agents shortest paths, importance ranking, corpus stats, confidence-filtered... | design |
| 2026-06-09 | Phase 2 Semantic Extraction (Ingester) Implementation Plan | This plan adds LLM-based semantic extraction to tatara-memory-repo-ingester. After the existing AST push, the ingester asks memory which analyzed files have changed (semantic-misses), chunks those files, calls... | - |
| 2026-06-09 | Phase 2: semantic ceiling (LLM extraction + graph analytics) | (operator wires the OpenAI key + semanticIngest flag into the ingest Job + the new CRD field) | design |
| 2026-06-09 | code_* Phase 2 MCP Tools Implementation Plan | This plan adds six new memory-target code_* MCP tools to internal/mcp/tools.go mapping to Phase 2 /code-graph/* routes, and adds an optional by param to the existing code_important tool. All new tools... | - |
| 2026-06-09 | Phase 2 Semantic Ceiling (tatara-memory) Implementation Plan | This plan implements the tatara-memory server-side slice of Phase 2: origin-scoped reconcile (extractor), the semantic cache (semantic_extractions), the Related/Hyperedge traversals, the in-process gonum... | - |
| 2026-06-09 | Phase 2 Semantic Ingest Wiring (tatara-operator) Implementation Plan | This plan wires the Phase 2 semantic ceiling into the operator's ingest path: a per-Repository semanticIngest opt-out (CRD field, default true), and three new env vars on the ingest Job's main container... | - |
| 2026-06-09 | SCM Projects CLI Tools Implementation Plan | Add 3 operator-mediated MCP intent tools to tatara-cli so agents can emit SCM intent (propose an issue, render a PR/MR review verdict, decide a self-authored PR/MR outcome) without ever calling SCM directly. Each... | - |
| 2026-06-09 | SCM-projects contract lock (shared wire types) | Purpose: the exact new types/endpoints/tool-schemas that tatara-operator (producer of REST + CRD), tatara-cli (consumer via MCP), and the SCM clients must match byte-for-byte. Locked before the per-repo plans so they... | - |
| 2026-06-09 | SCM Projects Infra/Deploy RUNBOOK Implementation Plan | Stand up the SCM-Projects feature in prod: a dedicated tatara-bot GitHub + GitLab identity with a per-Project fine-grained PAT, the board + approval label, the scm block patched onto each Project CR via the... | - |
| 2026-06-09 | SCM Projects, Proactive Issues, and PR/MR Reactions (tatara-operator) Implementation Plan | Extend the operator's SCM egress so tatara participates on a project board (GitHub Projects v2 / GitLab issue board), proactively opens human-gated issues (awaiting-approval), and reacts to PRs/MRs conditionally on... | - |
| 2026-06-09 | SCM Projects, proactive issues, and PR/MR reactions | (agent reach). Infra: tatara-bot SCM account + per-Project fine-grained PAT. | design (approved in brainstorm 2026-06-09) |
| 2026-06-09 | SCM Projects Wrapper Tool Flow-Through Implementation Plan | Make the three new SCM-projects MCP tools (propose_issue, review_verdict, pr_outcome) reach agents running inside the tatara-claude-code-wrapper, and prove they do. The wrapper does NOT enumerate MCP tools:... | - |
| 2026-06-09 | Server-side graphify: parity roadmap (strictly better, never worse) | Acceptance bar: tatara must be strictly better than graphify and in no case worse -- every graphify capability and data-model element matched or exceeded, no feature regressed. Parity is the floor; tatara's... | roadmap (organizing doc; per-phase specs follow) |
| 2026-06-11 | Autonomous Cron: tatara-cli issue_outcome MCP Tool Implementation Plan | Add the issue_outcome MCP tool to tatara-cli's OperatorTools() (operator tool count 12 -> 13), wiring it to POST /tasks/{TASK_ID}/issue-outcome, and bump the cli version 0.5.0 -> 0.6.0. | - |
| 2026-06-11 | Autonomous Cron - Contract Lock | This freezes every cross-repo wire type before parallel build, so the operator, cli, wrapper, and infra plans cannot diverge. Byte-for-byte. Field names, json tags, enum strings, route paths, label keys are normative. | Frozen wire types. Change here = change in all four repos. |
| 2026-06-11 | Autonomous Cron: MR/Issue Triage + Self-Driven Brainstorming - Design | Make tatara act autonomously on a schedule, not only in reaction to webhooks: | Design (approved, pre-plan) |
| 2026-06-11 | Autonomous Cron Infra/Deploy RUNBOOK Implementation Plan | Roll the autonomous-cron operator (new appVersion, carries ScmCron/BrainstormActivity/priorityLabel/Last*Scan CRD additions, the issue_outcome REST endpoint, the scan loop, and the `tatara.io/egress:... | - |
| 2026-06-11 | Autonomous Cron - tatara-operator Implementation Plan | Contract lock:** docs/superpowers/specs/2026-06-11-autonomous-cron-contract-lock.md (FROZEN wire types - field names, json tags, enum strings, route paths, label keys are byte-for-byte normative) | - |
| 2026-06-11 | Autonomous Cron - tatara-claude-code-wrapper Implementation Plan | Build order:** step 3 of 4 (operator -> cli -> wrapper -> infra). Depends on the cli having shipped issue_outcome in a new tagged image. | - |
| 2026-06-11 | Wrapper Transcript Streaming - Design | Emit every claude-code message and action (everything in/out: user prompts, assistant text, tool_use, tool_result, thinking) to the wrapper pod's stdout in real time, as structured JSON logs, so an operator watching... | Design (approved, pre-plan) |
| 2026-06-12 | Bot MR auto-merge on green pipeline - design (2026-06-12) | when the operator opens a bot MR and its pipeline succeeds, the forge | - |
| 2026-06-12 | Bot MR auto-merge on green pipeline - Implementation Plan | When the tatara operator opens a bot-authored PR/MR under mergePolicy: autoMergeOnGreenCI, enable the forge's native auto-merge so it merges to the default branch once required checks pass, and the linked issue closes. | - |
| 2026-06-12 | CI Pipelines for tatara Go Services - Design | Scope: subsystem A of the "CI pipelines + agent-wait-for-merge" feature. Subsystem B (operator waits for its PR to merge + main CI green, resolves conflicts, closes the associated issue) is a separate spec that... | design + as-built (subsystem A implemented) |
| 2026-06-12 | CI Pipelines (Subsystem A) Implementation Plan | Give each of the five tatara Go service repos a GitHub Actions pipeline (secscan/lint/test/build/smoke on PR; same plus a kaniko image build+push to harbor on push-to-main), running on the org-scoped self-hosted ARC... | - |
| 2026-06-12 | Issue-lifecycle agent - design | 2026-06-12). Supersedes the agent-side Issue/MR management handoff (docs/superpowers/handoffs/2026-06-12-agent-issue-mr-management.md) and the deferred subsystem-B "operator waits for its PR + closes the issue" scope. | - |
| 2026-06-12 | Issue-lifecycle M0 - lifecycle model + skeleton - Implementation Plan | Introduce the issueLifecycle Task kind, its Status.LifecycleState | - |
| 2026-06-12 | Issue-lifecycle M1 - D babysit (MRCI / Merge / MainCI / give-up) - Implementation Plan | Implement the poll-and-react babysit states so a bot PR drives itself | - |
| 2026-06-12 | Issue-lifecycle M2 - conversation (Triage discuss / Conversation idle) - Implementation Plan | Close the conversation loop: the agent posts questions (discuss, wired | - |
| 2026-06-12 | Issue-lifecycle M3 - context-guard handover - Implementation Plan | When an S7 (MainCI-failure) re-implement would run with a near-full | - |
| 2026-06-12 | Issue-lifecycle M4 - MR scope completeness - Implementation Plan | The Implement run reports what it delivered; the MR body states the | - |
| 2026-06-12 | Per-Project External Exposure (memory + chat) - Design | When a Project CR is reconciled, the tatara-operator provisions and exposes that project's per-project memory (already provisioned) and chat (newly provisioned) via a per-project Ingress on the shared host,... | design, pending user review |
| 2026-06-12 | Per-Project External Exposure Implementation Plan | The tatara-operator provisions and exposes per-project memory (phase 1) and chat (phase 2) via a per-project Ingress on the shared host, so the tatara MCP tools reach the project's code graph externally with... | - |
| 2026-06-12 | Per-repo all-items fan-out - design (2026-06-12) | every open issue/MR in every project repo gets triaged, parallelised by | - |
| 2026-06-12 | Per-repo all-items fan-out - Implementation Plan | Make the autonomous cron triage every open issue/MR in every project repo, parallelised per repo (one lane per repo, maxPerRepo=1 active Task), bounded by the existing MaxConcurrentTasks ceiling. | - |
| 2026-06-13 | Per-repo backpressured brainstorm + approval backstop | Two gaps in the autonomous scan loop: | approved (design), pending implementation |
| 2026-06-13 | Brainstorm Backpressure + Approval Backstop Implementation Plan | Make the operator's brainstorm activity per-repo and backpressured (hourly; skip a repo once it holds >= maxOpenProposals open unapproved agent proposals; else start one brainstorm), and add an approval backstop that... | - |
| 2026-06-13 | Discovery-phase research harness - design | Spec owner repo: tatara (docs index) | approved (brainstorm) - ready for plan |
| 2026-06-13 | Discovery-phase research harness Implementation Plan | Ship two wrapper agent skills (tatara-deep-research, tatara-research-followup) plus the MCP/operator plumbing they need, so the autonomous loop's discovery phase does deep cross-platform research and keeps... | - |
| 2026-06-13 | infra-helmfile: tatara-helmfile Deploy Runner + tatara Bucket Removal Implementation Plan | In the existing GitLab ~/Documents/infra/helmfile repo, add a dedicated in-cluster ARC runner scale set (arc-runner-tatara-helmfile) that runs as a new cluster-admin-bound ServiceAccount for autonomous `helmfile... | - |
| 2026-06-13 | tatara-helmfile extraction + autonomous deploy harness - design | Stand up a dedicated private tatara-helmfile GitHub repo that owns the tatara platform's helm releases, deploys them to the cluster via GitHub Actions on self-hosted in-cluster runners, is enrolled as a... | approved (user pinned all decisions upfront, authorized run-to-completion) |
| 2026-06-13 | tatara-helmfile Repo Implementation Plan | Stand up a standalone private tatara-helmfile repo that owns the two tatara helm releases, deploys them to the cluster via GitHub Actions on an in-cluster ARC runner with an in-cluster ServiceAccount, and... | - |
| 2026-06-13 | tatara-managed 3-label issue lifecycle - design | Today the operator exposes three separate, inconsistent human-signal surfaces: | approved (design), pending spec review |
| 2026-06-13 | tatara-managed 3-label issue lifecycle - Implementation Plan | The operator manages exactly one of tatara-idea / tatara-approved / tatara-rejected on every issue in a managed repo, driven by the triage agent reading the conversation; the label-toggle approval subsystem is... | - |
| 2026-06-13 | tatara phase-label dedup + orphan-recovery - design | The original mechanism - dedup/orphan-recovery by label-added-time (T_L) vs Task-creation-time plus a new SCMReader.IssueLabelAddedAt - is WITHDRAWN. It cannot work: the operator sets a phase label DURING a... | AMENDED 2026-06-13 (Option A) - ready for implementation |
| 2026-06-13 | tatara phase-label dedup + orphan-recovery (Option A) - implementation plan | Make the issue's phase label the lifecycle state-of-truth and kill | - |
| 2026-06-13 | tatara-deploy-harness Skill (Sub-system D) Implementation Plan | Ship a baked tatara-deploy-harness skill in tatara-claude-code-wrapper that drives the 9-step S1..S9 in-session deploy state machine (research -> implement -> component MR -> helmfile MR -> apply -> deliver) with... | - |
| 2026-06-14 | BuildKit-on-Ceph Migration Implementation Plan | Replace kaniko in-cluster image builds with a single rootless buildkitd | - |
| 2026-06-14 | Issue #34: Empty Implement Run -> Retry-then-Escalate Implementation Plan | An issueLifecycle Implement turn that produces zero commits must no longer silently park as benign no-change (fix never lands). Instead: retry up to 2x with a re-entry prompt, then comment on the issue and park... | - |
| 2026-06-14 | mise toolchain across tatara | (1) wrapper container ships mise; (2) each project has a .mise.toml and builds via mise; (3) CI uses .mise.toml; (4) agent prompt is mise-aware. All tested. | - |
| 2026-06-14 | pre-commit as the unified lint/test gate | (1) each repo runs pre-commit hooks (lints at commit stage, tests at pre-push stage); (2) the wrapper installs the hooks after cloning so the agent's commits/pushes are gated; (3) CI runs pre-commit run instead of... | - |
| 2026-06-14 | Wrapper: detect claude-code death and resume the session | When the claude-code process dies, the wrapper must detect it immediately (not wait for the 1800s turn-timeout), relaunch claude with --continue to restore the conversation, and resume the in-flight turn by... | - |
| 2026-06-15 | Auto-close recovery-exhausted bot PRs — Design + Plan | When the revived recovery cannot land a bot PR after maxRecoveryAttempts adoption cycles, mrScan CLOSES the PR (branch preserved, reopen-to-retry comment) instead of leaving it parked open. Closes the egress gap... | - |
| 2026-06-15 | Brainstorm dedup + comment-on-existing-issue — Design | The hourly brainstorm agent opens a new issue every cycle (propose_issue is a hard requirement) with no real dedup: the goal delegates dedup to the tatara-deep-research skill, which was never built, and the agent... | approved design (scope/mechanism/context settled), pre-plan |
| 2026-06-15 | tatara platform - deep audit report | Generated 367 raw findings across 32 audit units (6 Go services + Argo CI + cross-cutting) via a 3-stage workflow (115 agents): deep finder -> adversarial verifier (file:line confirmed) -> independent skeptic for... | - |
| 2026-06-15 | Implement-refusal codification — design | Make "the agent declined to implement" a first-class, codified outcome. | - |
| 2026-06-15 | Revive & harden the bot-PR merge/recovery pipeline | Five bot-authored PRs sat unmerged indefinitely: operator#50/#41/#39 (CONFLICTING), wrapper#25 (MERGEABLE), chat#31 (CONFLICTING). The autonomous system both let them go AND never recovered them. | approved design, pre-plan |
| 2026-06-15 | Revive & Harden Bot-PR Merge/Recovery — Implementation Plan | Restore the operator's autonomous bot-PR recovery (dead due to a laneOccupancy accounting bug), bound it against thrash, quiet a cosmetic reconcile log, deploy via GitOps, and let the revived system heal the 5... | - |
| 2026-06-16 | tatara platform - deep audit report (ROUND 2) | Post-remediation round-2 audit (after round-1 merged 365 fixes). 34 units, 3-stage find->verify->skeptic. 269 confirmed (regressions, missed/deferred, new defects). | - |
| 2026-06-16 | Project-level brainstorm + healthCheck (drop repo pinning) | brainstorm and healthCheck are already PROJECT-scheduled (one Task per cycle, project-wide in-flight guard + proposal-backlog cap) but each Task is pinned to a deterministic primaryRepo (`TaskSpec.RepositoryRef... | - |
| 2026-06-19 | Brainstorm: no repeat-comment, prefer novelty — Design | The hourly project-wide brainstorm agent surveys open issues and, per the 2026-06-15 brainstorm-dedup design, may call comment_on_issue when its idea "connects to" an existing issue instead of opening a duplicate.... | approved design, pre-plan |
| 2026-06-19 | Brainstorm: no repeat-comment, prefer novelty — Implementation Plan | Stop the project-wide brainstorm/healthCheck agent from commenting on the same issue every cycle: flag issues the bot already commented on, steer the prompt to novel work, and hard-gate a duplicate bot comment at the... | - |
| 2026-06-19 | Grafana incident-response — Design | single milestone, end-to-end. Optional per-project feature. | approved design, pre-plan |
| 2026-06-19 | Grafana incident-response Implementation Plan | Optional per-project Grafana incident-response: receive a Grafana alert webhook, provision a read-only grafana-mcp, run an incident agent that investigates live and files an evidence issue the brainstorm/lifecycle... | - |
| 2026-06-20 | Agent-work queue (in-operator) - Design | single milestone, end-to-end. Per-project, opt-out via defaults that match today's behaviour. | approved design, pre-plan |
| 2026-06-20 | Agent-work Queue (in-operator) Implementation Plan | Fold a durable, ordered admission queue into tatara-operator so every agent-spawning path enqueues a QueuedEvent instead of creating a Task directly; a dispatcher admits them in strict global seq FIFO under... | - |
| 2026-06-20 | Cross-repo-aware brainstorm + ultracode agents - design | The autonomous brainstorm/implementation loop is repo-myopic and produces low-grade artifacts: | approved (brainstorm), plan + implementation pending |
| 2026-06-20 | Plan: brainstorm-systemic - tatara-cli (WS3, cli half) | Spec: docs/superpowers/specs/2026-06-20-brainstorm-systemic-and-agent-ultracode-design.md (WS3) | - |
| 2026-06-20 | Plan: Cross-repo-aware brainstorm + ultracode agents - tatara-operator | For agentic workers using superpowers:subagent-driven-development. Each Task below is a self-contained TDD unit: dispatch one sonnet implementation subagent per Task with this file's path + the Task number. Tasks... | - |
| 2026-06-20 | Plan: brainstorm-systemic - WS2 agent ultracode (wrapper) | Spec: docs/superpowers/specs/2026-06-20-brainstorm-systemic-and-agent-ultracode-design.md (WS2) Sibling plans: ...-operator.md, ...-cli.md | - |
| 2026-06-22 | Brainstorm throughput + queue-as-concurrency | Brainstorm runs hourly per project but delivers very few new issues. Root cause is three compounding throttles, evidenced live (Prometheus + Loki, 2026-06-22): | approved, ready for planning |
| 2026-06-22 | Brainstorm throughput + queue-as-concurrency Implementation Plan | Make hourly brainstorm actually deliver new issues by removing the redundant autonomous-enqueue cap (queue limits concurrency, not creation), raising maxOpenProposals to 10, focusing brainstorm on new ideas only,... | - |
| 2026-06-22 | Per-project guidance + tatara self-infra awareness | The tatara project is self-hosting: its repositories (operator, cli, wrapper, memory, ingester, helmfile) ARE the autonomous-agent platform the agents run on. Today brainstorm/healthCheck survey the repos for... | approved, ready for planning |
| 2026-06-22 | Per-project guidance + tatara self-infra awareness Implementation Plan | Add an optional spec.scm.guidance string to the Project CRD, append it to the brainstorm + healthCheck goal context, and set the tatara project's guidance so operational-infrastructure improvements are in-scope. | - |
| 2026-06-23 | Spec: agent MCP tool surface (per-phase gating + consolidation + report_internal_issue) | One coherent change to the MCP tool surface that tatara agent pods see, made of three parts: | design approved (decisions resolved 2026-06-23), ready for implementation |
| 2026-06-23 | Tatara Deep-Research Capability: Architecture Recommendation | 1. Egress: broad ipBlock: 0.0.0.0/0 (reuse existing deploy-sample policy). Accepted blast-radius: brainstorm-pod bot PAT + OIDC creds exfiltratable. Mitigation: keep search-API keys out until Phase 2;... | - |
| 2026-06-23 | Defect B - Duplicate issueLifecycle Tasks (operator) Implementation Plan | Stop the operator from spawning a second issueLifecycle Task for an issue that already has one (adopt + re-triage the existing Parked/terminal Task) and stop the operator's own bot comments from re-triggering a rescan. | - |
| 2026-06-23 | Defect C: false-refusal already_done + decline hardening Implementation Plan | Stop the operator falsely parking implement tasks as refused-no-explanation when the change is already present, by adding an already_done outcome action and forcing every decline/already_done to carry a... | - |
| 2026-06-23 | False-Refusal Wrapper Half (WT-3) Implementation Plan | In tatara-claude-code-wrapper, propagate a rejected critical-outcome MCP tool call (decline_implementation / already_done) back to the agent by re-prompting it instead of finishing the turn silently (Defect C... | - |
| 2026-06-23 | Cross-Repo Declarative Scope (ReposInScope) Implementation Plan | Make cross-repo implement scope declarative on the Task (optional Spec.ReposInScope), inject it into the implement prompt, and warn on the source issue when an in-scope repo produced no branch instead of silently... | - |
| 2026-06-23 | Design: Implement false-refusal root-cause fix (helmfile#8) | Deploy: GitOps via tatara-helmfile (cli -> wrapper -> operator) | approved, ready for planning |
| 2026-06-24 | Task CR as first-class lifecycle tracker (work-item ledger) | The Task CR is the first-class tracker for the whole brainstorm -> implement -> review lifecycle. It carries a ledger of every SCM work-item it spans (N tracker issues + N MRs + brainstorm proposals), which... | design - pending user review |
| 2026-06-24 | Task Work-Item Ledger Implementation Plan | Make the Task CR a Project-level first-class tracker carrying a work-item ledger (N issues + N MRs + proposals) that is the single source of truth for dedup, stall recovery, brainstorm triage, and agent... | - |
| 2026-06-24 | tatara Grafana alerting -> emergency-brainstorm - Design | single milestone, end-to-end. Builds on the shipped 2026-06-19 Grafana incident-response feature (docs/superpowers/specs/2026-06-19-grafana-incident-response-design.md). | approved design, pre-plan |
| 2026-06-24 | tatara Grafana alerting -> emergency-brainstorm Implementation Plan | Define Grafana-managed alerts (metric + Loki) for every tatara component, labelled system=tatara, so they route through the existing operator incident webhook into an emergency-brainstorm that files a... | - |
| 2026-06-25 | tatara-observability: agent-adjustable alerts-as-code - Design | tatara-helmfile (enroll). Status: approved design. | - |
| 2026-06-28 | Semver push-CD - cutover runbook | Covers the gated, one-time, mostly out-of-repo steps the implementation workflow could NOT perform (tag pushes, repo settings, Harbor policy, secrets, the enable-auto-merge cutover). The code lives on... | SHIPPED 2026-07-05 |
| 2026-06-28 | Semver push-CD for tatara | cross-repo. Per-component implementation specs are produced during writing-plans. Originating issue: szymonrychu/tatara-helmfile#87. | SHIPPED 2026-07-05 |
| 2026-07-04 | Cheap Adjacents (Token Conservation, Component 5) Implementation Plan | land the four independent, low-effort, high-ROI token-conservation | - |
| 2026-07-04 | Claude subscription usage gate: complete claudeSubscription into a proactive per-kind spawn gate | The tatara fleet burned the Claude subscription too fast and had to be scaled to 0. The whole fleet authenticates with ONE shared CLAUDE_CODE_OAUTH_TOKEN (a personal Max-plan subscription OAuth token,... | deploy-gated |
| 2026-07-04 | Claude Subscription Usage Gate Implementation Plan | Complete the merged-but-inert claudeSubscription budget mode into a working proactive per-kind spawn gate driven by a standalone operator-side poller of Claude's account usage windows. | deploy-gated |
| 2026-07-04 | Durable token/cost measurement (G2 + G12) | tatara-operator, tatara-observability (+ tatara-claude-code-wrapper only if the turn-complete callback must forward cache_creation) Origin: token-conservation P0 review gaps G2 + G12... | design (approved, pre-plan) |
| 2026-07-04 | Implementation Plan: Durable token/cost measurement (G2 + G12) | Spec: docs/superpowers/specs/2026-07-04-durable-measurement-design.md | - |
| 2026-07-04 | Handoff-based continuation (replacing S3 conversation restore) | tatara-chat, tatara-cli, tatara-claude-code-wrapper, tatara-agent-skills, tatara-operator Origin: token-conservation component 3 (handoff-replaces-S3), reshaped by the user's directive that stale-session S3 restore... | design (approved-by-delegation; user granted full autonomy) |
| 2026-07-04 | Handoff-based continuation - Implementation Plan | Replace S3 conversation restore with compact handoffs stored in tatara-chat, queryable + groomable by the platform's own agents. | - |
| 2026-07-04 | Per-kind Model + Effort tiering Implementation Plan | Attack axis 1 of the token-conservation spec (component 1). Add | - |
| 2026-07-04 | Quality-feedback loop (G4 + G5) | tatara-operator, tatara-observability Origin: token-conservation P0 review gaps G4 + G5 (docs/2026-07-04-token-conservation-p0-review-gaps.md). G6 ($-budget) is a separate follow-on spec. | design (approved, pre-plan) |
| 2026-07-04 | Quality-feedback loop (G4+G5) Implementation Plan | Instrument the Sonnet-review/triage downgrade with model-keyed quality-proxy metrics (G4) and close a self-tuning loop that auto-proposes a tier-revert MR on regression (G5). | - |
| 2026-07-04 | Token conservation redesign | platform-wide (tatara-operator, tatara-cli, tatara-claude-code-wrapper, tatara-agent-skills, tatara-observability, tatara-helmfile) | deployed |
| 2026-07-04 | Token-conservation P0 review: verdict + gaps for brainstorm | review agents: operator diff, wrapper+observability diff, helmfile deploy branch, design-level gap analysis) plus live Prometheus checks. | deployed |
| 2026-07-04 | Token-conservation fleet re-enable runbook (G3) | The agent fleet was scaled to 0 to stop token burn. P0 tiering + adjacents, the G1/G2/G12 durable measurement, and the G7/G15/G17 hardening are all deployed. This runbook brings the fleet back staged, not... | deployed |
| 2026-07-04 | Token Measurement + Observability Implementation Plan | Component 6 (Measurement) of the token-conservation redesign (tatara/docs/superpowers/specs/2026-07-04-token-conservation-design.md). This is the prerequisite instrument and the gate for component 4 (cache... | - |