2026-07-24 - 2026-10-24
Overview
1 Release published by 1 user
Published
v0.1.0
v0.1.0
136 Pull requests merged by 1 user
Merged
#179 The call record: every model call, for the operator's study (M6.6a)
Merged
#180 Benchmark Qwen3.8-Flash-Next against qwen3.8-27b
Merged
#178 Docs: the status page and the call record (#109)
Merged
#177 Benchmark: workspaces are git repositories
Merged
#176 Docs: identity is Samba AD with Authelia, not LLDAP
Merged
#175 Log every model call's time and tokens
Merged
#171 M7.6: nisse-review's image
Merged
#169 nisse-review marks a reply that settles its thread; the PR's author resolves it
Merged
#170 CI: one e2e at a time across every PR
Merged
#166 The agent host's vault and SilverBullet (M5.4)
Merged
#168 nisse-review: a review refused for an open point posts once the point is answered
Merged
#167 Development Plan: every agent is its own container image
Merged
#165 A model's safety-filter refusal is an error that says so, not an empty reply
Merged
#164 Trust test: wait for Gitea to dismiss a stale approval before merging (#162)
Merged
#163 Compaction: the model's notes replace the middle of a long run (#158)
Merged
#160 nisse-review: read code from a git clone kept between runs
Merged
#159 nisse-review: a stop ends a review in progress; quieter repo failures
Merged
#161 nisse-dev: a VM lost as a helper's work is taken back ends the run
Merged
#156 nisse-review: later reviews, replies, rounds, stop and review (M7.5)
Merged
#157 nisse-review: read long files in parts; a retry starts afresh
Merged
#155 nisse-review: a new head's review waits for its CI
Merged
#154 nisse-review: match repository patterns against Gitea's search
Merged
#153 nisse-review: repository patterns, and the wall checked every round
Merged
#152 CI runs nisse-review's live test
Merged
#151 M7.4: nisse-dev answers a peer
Merged
#149 M7.3: nisse-review's first slice
Merged
#148 M7.2: share the loop's parts
Merged
#147 nisse-core: group modules into packages
Merged
#146 Spec: nisse-review (Part R) before nisse-ops
Merged
#145 Architecture X3: avoid, not reject, search-then-dispatch meta-tools
Merged
#143 M7.1: extract nisse-core
Merged
#142 M5.3: scope memory
Merged
#141 M5.2: agent memory
Merged
#140 M5.1: the knowledge vault
Merged
#139 M4.4: an unreachable build VM blocks its projects until it's back (#137)
Merged
#138 M4.3: a broken build VM blocks its projects until it's ready
Merged
#136 M4.2: the toolchain step
Merged
#135 Architecture: the toolchain step (D3), and only looking before the go-ahead (D1)
Merged
#134 Expert and integration roles on claude-haiku-5-5
Merged
#133 M4.1: toolchain preflight
Merged
#132 Say a local commit publishes nothing (#129)
Merged
#131 M3.3: the trust dial (the agent merges approved PRs)
Merged
#128 M3.2: the action classifier
Merged
#127 M3.1: action classes and the fast path
Merged
#126 CI: a newer run cancels the one it supersedes
Merged
#125 M2.8: switch over to the new loop
Merged
#122 Fixes from the rest of the live session
Merged
#119 Fixes from the first live session of the new loop
Merged
#118 M1.5 accepted: tests over a published image
Merged
#117 Test levels: smoke e2e on PRs, full set on e2e:full and before a release
Merged
#116 Wait out LLM outages with backoff (L5)
Merged
#115 The agent as a container image (M1.5)
Merged
#114 Conversation on the new loop (M2.7)
Merged
#113 Roles, capability profiles and help on the new loop (M2.6)
Merged
#108 New projects from conversation (M2.5, PR B)
Merged
#110 End-to-end tests on a local model, and in CI on every PR
Merged
#107 A repository per test run instead of the shared sandbox
Merged
#106 Conversations on the new loop (M2.5, PR A)
Merged
#105 Projects as configuration entries (M1.4)
Merged
#104 Architecture: the general scope and the new-project sequence (for M2.5)
Merged
#103 Multi-milestone work and documents on the new loop (M2.4)
Merged
#102 Threads, commands and labels on the new loop (M2.3)
Merged
#101 The queue: parking, the operator's order, blocked items (M2.2)
Merged
#100 Answer where asked, wait on a merge, and let the driver title PRs
Merged
#99 Properties against the real loop, and the slice end to end (M2.1, PR D)
Merged
#97 The new loop's core: intake, restatement, go-ahead, threads (M2.1, PR B)
Merged
#96 Matrix threads in the transport (M2.1, PR A)
Merged
#98 Milestones and review in the new loop (M2.1, PR C)
Merged
#94 Mark M1.3 done in the Development Plan
Merged
#95 Add the end-to-end test, run after each milestone (M1.6)
Merged
#93 Run every command on the build VM over SSH (M1.3)
Merged
#92 Rework the Development Plan around live testing
Merged
#91 Model the agent loop as the architecture describes it
Merged
#90 Revise the deployment model: an agent host and tool VMs
Merged
#88 Give each role its own view of the transcript (M1.2)
Merged
#87 Work items in the state store, with a transcript bound to each (M1.1)
Merged
#86 Add a bring-up script and a transcript dump; let nisse-dev install on its own VM (M0.2)
Merged
#85 Retire the old build spec, and point every reference at the architecture (M0.1)
Merged
#84 Propose the Development Plan
Merged
#83 Draft: rewrite the architecture (docs/Architecture.md)
Merged
#76 Redeliver an interrupted review comment, and close the issue on completion
Merged
#75 Adopt an already-published PR instead of publishing again
Merged
#74 Bind approvals and feedback to the revision they were about
Merged
#73 Model the orchestrator in Quint, and reproduce its counterexamples
Merged
#72 Match approvals and commands regardless of case and trailing punctuation
Merged
#71 Add the abandon command
Merged
#68 Keep Gitea threads watched across restarts and interruptions
Merged
#67 Don't assert the exact word in the LiteLLM smoke test
Merged
#66 Never leave the agent stranded in a working state
Merged
#63 Publish work the executor committed itself, and gate when there is none
Merged
#60 Handle leading mentions in Matrix messages
Merged
#57 Keep a pending question across restarts, and make the restart question answerable
Merged
#56 Use qwen3.8-27b as the example executor
Merged
#55 Benchmark the finalists on all eight tasks, and archive older results
Merged
#54 Spec: prefer one local model for all roles; record host sharing, park vision
Merged
#53 Add four harder benchmark tasks
Merged
#52 Keep runs with different tuning apart in the report, and raise max_turns to 40
Merged
#51 Render one assistant message per turn, and make the request timeout a setting
Merged
#50 Record rate-limit waits in benchmark runs, and leave them out of time
Merged
#49 Benchmark seven new RTX 4090 models, and local-fast on the 4090
Merged
#48 Point the example config at Opus 5.5, with Haiku as consultant
Merged
#47 Add an executor benchmark that keeps its results
Merged
#45 Enforce the tool-call contract, asking servers only where it works
Merged
#44 Fit every model request to its context window
Merged
#43 Move configuration to YAML, and make the tuning knobs configurable
Merged
#41 Add the operator kill switch, stop (A4)
Merged
#39 Gate ops writes on approved step sequences
Merged
#38 Define the work model: duties, work items, scopes
Merged
#36 Record the MCP server deployment decisions in the spec
Merged
#35 Discover a model's vendor from LiteLLM instead of its name
Merged
#34 Choose the LLM route by model, and add a capability matrix
Merged
#31 Add the composition root and a runnable entry point
Merged
#25 Wire delegation to A12's trigger, and implement A3's resume rule
Merged
#24 Add ask_expert and eval capture (A14/A16) — step 7c
Merged
#23 Map A17 onto Outline's actual primitives, and record the MCP decision
Merged
#22 Add A17, the knowledge store contract (Outline)
Merged
#21 Add the delegation harness (A13) — step 7b
Merged
#20 Add the Anthropic passthrough renderer and client (A10/A13) — step 7a
Merged
#19 Record default-branch protection as a B2 requirement
Merged
#17 Add PR comment intake and the review loop (B2) — step 6b
Merged
#18 Externalise plans to Gitea and accept approvals there (A4, Part H)
Merged
#16 Publish finished milestones as PRs (B2/A3) — step 6a, part 2
Merged
#15 Add a manual CI trigger (workflow_dispatch)
Merged
#12 Add executor tool loop (A12/A15) — step 5b
Merged
#14 Add Gitea client and git operations (B2) — step 6a, part 1
Merged
#13 Assert the live Matrix tests run as a testing identity
Merged
#11 Add LLM-backed planning agent (A4/A11/A13) — step 5a
Merged
#10 Wire live-test credentials into CI from Actions secrets
Merged
#9 Add MatrixTransport (A1) — step-2 PR B
Merged
#8 Add LLM client interface with LiteLLM adapter (A5/A11)
Merged
#7 Add canonical event log and renderer skeleton (A10)
Merged
#6 Add control loop, gating primitives, and transport interface (A4/B1)
Merged
#5 Implement A9 state-change logging in the state store
Merged
#4 Amend A9 with state-change log conventions
Merged
#2 Add durable state store and boot-time reconciliation (A2/A3)
Merged
#1 Scaffold hello-world apps, test frameworks, and Gitea CI
21 Issues closed from 2 users
Closed
#162 Flaky: test_merging_needs_the_operators_approval_of_the_pr_as_it_is (Gitea let a stale approval merge)
Closed
#158 Context compression: fit the window without losing what matters or the prompt cache
Closed
#137 Block projects on an unreachable build VM, and unblock when it's back
Closed
#129 Driver commits by hand and claims a PR exists
Closed
#65 Let the agent merge its PRs once the operator has approved them
Closed
#80 The publisher commits build artifacts into PRs (git add -A)
Closed
#89 Gitea token is stored in the workspace clone's .git/config
Closed
#78 The planner's transcript: scope it to the task, attribute who acted, and include orchestrator facts
Closed
#33 Add ability to reference Vikunja tickets with agent accounts.
Closed
#30 Idea: one agent instance per project, with automated provisioning
Closed
#61 Simplify milestone gating: start on approval, and skip the milestone gate for single-milestone projects
Closed
#79 The planner can't say 'no more milestones', and 'next milestone' forces one
Closed
#81 A milestone that legitimately changes nothing has no way forward
Closed
#46 Separate architect and planner roles
Closed
#82 Discussion: architecture as a project document, separate build plans, and how detailed plans should be
Closed
#64 Approvals and operator commands shouldn't be case sensitive
Closed
#58 Restart in PR_OPEN asks 'continue or re-plan?' although nothing was in flight
Closed
#59 Boot doesn't re-attach the Gitea plan and review pollers
Closed
#32 Move configuration from .env to a YAML config file
Closed
#40 Add the operator kill switch, stop (A4)
Closed
#3 Define clear logging for state changes.
44 Issues created by 2 users
Opened
#3 Define clear logging for state changes.
Opened
#26 Future: multiple forges and multiple projects (GitHub alongside Gitea, several Gitea instances)
Opened
#27 Per-project state and clone-on-use (A2/B1 debt)
Opened
#28 Room per project, created by the agent
Opened
#29 Create a project from scratch, on approval (blocked on an ownership decision)
Opened
#30 Idea: one agent instance per project, with automated provisioning
Opened
#32 Move configuration from .env to a YAML config file
Opened
#33 Add ability to reference Vikunja tickets with agent accounts.
Opened
#37 Add support for MCP servers, including in configuration file.
Opened
#40 Add the operator kill switch, stop (A4)
Opened
#42 Post the model's per-turn commentary, and show delegation progress
Opened
#46 Separate architect and planner roles
Opened
#58 Restart in PR_OPEN asks 'continue or re-plan?' although nothing was in flight
Opened
#59 Boot doesn't re-attach the Gitea plan and review pollers
Opened
#61 Simplify milestone gating: start on approval, and skip the milestone gate for single-milestone projects
Opened
#62 Create the milestone branch when execution starts, not at publish
Opened
#64 Approvals and operator commands shouldn't be case sensitive
Opened
#65 Let the agent merge its PRs once the operator has approved them
Opened
#69 Inline PR review comments are never seen
Opened
#70 Re-attaching a watch from inside its own callback stalls ~25 s
Opened
#77 Make the agent conversational: questions, explanations and asking back, in Matrix
Opened
#78 The planner's transcript: scope it to the task, attribute who acted, and include orchestrator facts
Opened
#79 The planner can't say 'no more milestones', and 'next milestone' forces one
Opened
#80 The publisher commits build artifacts into PRs (git add -A)
Opened
#81 A milestone that legitimately changes nothing has no way forward
Opened
#82 Discussion: architecture as a project document, separate build plans, and how detailed plans should be
Opened
#89 Gitea token is stored in the workspace clone's .git/config
Opened
#109 Consider using a shared DB between all agents along with a UI to show status/progress and backlog for all agents.
Opened
#111 Let's have the dev agent review new issues.
Opened
#112 Issue labels are not free form.
Opened
#120 Link issues, PRs and threads in Matrix messages
Opened
#121 End the closing message with a clear, consistent last line
Opened
#123 Create additional agents
Opened
#124 Consider additional conversational transports
Opened
#129 Driver commits by hand and claims a PR exists
Opened
#130 Project setup script (protection, labels) and a label check
Opened
#137 Block projects on an unreachable build VM, and unblock when it's back
Opened
#144 Redact secrets an MCP server returns, not only the credentials we hand out
Opened
#150 Use whatmodel.sh to suggest model updates
Opened
#158 Context compression: fit the window without losing what matters or the prompt cache
Opened
#162 Flaky: test_merging_needs_the_operators_approval_of_the_pr_as_it_is (Gitea let a stale approval merge)
Opened
#172 Flaky: test_git ssh-agent tests count every ssh-agent on the box
Opened
#173 nisse-review: approve PRs where its approval counts
Opened
#174 nisse-review: a timed-out model call shouldn't throw away the whole review