| Commit message (Collapse) | Author | Age | Files | Lines |
|---|
| |
|
|
|
|
|
|
| |
All six decisions resolved. The router's input is filed keepers that belong to another project, not raw inbox files (Reading B). That keeps it a separate sub-step from the inbox gate (D1) and distinct from the defer-and-stage router (D5). Transcript routing is deferred to vNext (D4).
I reworked the design to match: the input definition, a candidate-set note bounding the router to session-filed keepers rather than the standing backlog, and Phase 3. The cookie reads [6/6] and the Status moved to ready for review.
The A-vs-B input ambiguity was the root under D1 and D5. Reading B keeps the inbox gate, the router, and defer-and-stage each simple instead of entangling all three.
|
| |
|
|
|
|
|
|
| |
A spec for the optional wrap-up step that routes inbox items (and, vNext, transcripts) to the project they belong to. Three decisions settled from grounding (reuse todo-cleanup's Open Work matcher as the destination anchor, move atomically through one helper, keep cross-project writes visible with a provenance note); three left open for Craig (separate router step vs merged into the inbox sanity check, transcript scope and trigger, reconciling with the defer-and-stage router). Five implementation phases, acceptance criteria, readiness dimensions.
Status stays draft while decisions are open. The todo task moves to DOING and links the spec.
From the archsetup handoff 2026-06-13.
|
| |
|
|
|
|
| |
A wrap-up router that surfaces outstanding inbox items, recommends a destination project for each, and batch-moves task items into that project's todo.org, with a parallel transcript-filing step. I captured it as a design source and filed a spec-bound feature task rather than building it now: the work clears the spec bar, with design uncertainty (merge or stay separate from the inbox sanity check, recommendation-engine confidence, an unresolved transcript source-location dependency) and overlap with this session's defer-and-stage router to reconcile.
From the archsetup handoff 2026-06-13.
|
| |
|
|
| |
The re-review confirmed every disposition with no new high or medium findings: Phase 1.5 stands at Ready with caveats, phases 2-5 stay parked behind the decisions fence. The response is correspondingly small — the accepted editorial rename of the Emacs subsection (its "open issue / blocks readiness" heading outlived the body, which is now an integration contract) and the second-pass note in the dispositions section. The updated review file and its history and task-tracking edits ride along.
|
| |
|
|
|
|
| |
The review's top finding was that one Not-ready label hid an implementable slice. Status now splits by arc: Phase 1.5 helper instances are READY WITH CAVEATS (the three-ring gate and the manual drills are binding, and the ai-term.el work is a coordinated .emacs.d handoff with an exact artifact), while phases 2-5 stay NOT READY behind a decisions-required section and a Phase 5 reverification prerequisite that demotes the model table to a recommendation.
The remaining findings hardened the slice: per-ring rollback actions including the half-propagated-sync case, the review's test inventory adopted as normative, a message contract for stale helper files, and explicit roster-unavailable behavior on unsupported platforms. All recommendations accepted except the document split, modified to a dual rubric in one document. The review file and dispositions table ride along.
|
| | |
|
| |
|
|
|
|
| |
The detection and identity subsections were authored under one design and patched under two others. They're now three pieces: the roster as the single detection primitive, two spawn paths (deterministic launcher, startup safety net) sharing the same three steps, and identity plus the role contract, with helper-mode.org named the single home of the helper rules.
Verified ai-term.el's launch mechanics in code: the startup instruction is embedded in the tmux new-session command string, so env injection is a prefix on that string, and the -A flag means one session per project, so the helper path needs its own session and buffer names. Added ai-term recommendations to ride the same handoff (roster-derived badges, agent-exit visibility, one source for opener strings) and reconciled the v0 session-context symlink question with the Phase 1.5 answer.
|
| |
|
|
|
|
| |
The first draft of the Emacs open issue assumed eat/vterm and invented an ai --no-tmux mode for a tmux-less path that doesn't exist. Verified against the actual config: the terminal is ghostel (native emulator over libghostty-vt), and ai-term.el is already the Emacs AI launch surface, creating project-named aiv- tmux sessions with liveness badges and crash recovery.
Emacs-born agents are therefore tmux-parented like shell launches, so detection is uniform across surfaces. The remaining design is integration, not a new surface: ai-term.el's session-create learns the roster, export, and opener steps, the picker gains a [helper] badge, and the launchers share only the agent-roster script since ai-term owns its own session naming and window placement.
|
| |
|
|
|
|
|
|
| |
gating
Two gaps block implementation. Sessions are also born from Emacs terminal buffers, where roster detection works (the scan matches process cwd, and eat/vterm shells are children of emacs) but the deterministic spawn path doesn't exist; the open issue weighs an elisp command against shelling out to ai with a no-tmux mode, leaning to the latter so the logic lives once.
Second, template sync makes "live everywhere" the default failure mode for startup.org changes, so the test strategy gains three-ring gating: bats with sleeper processes and a byte-identical no-op guarantee, a disposable sandbox project for the corruption, orphaned-helper, and raw-launch drills, then a dormant-by-construction pilot through project-scripts before the template-wide release. The Status section carries the readiness checklist and the implementation task is blocked on it.
|
| |
|
|
|
|
| |
The launcher becomes the spawn mechanism: a shell script runs the roster check, assigns the id, and launches with the helper instructions in order, where a model-followed startup instruction can skip a step. The in-session roster check stays as the safety net for raw launches and still splits a live anchor into crashed versus concurrent.
Session-end ordering was unhandled: a helper outliving the primary stranded a dirty worktree, since the helper may not commit and the agent allowed to is gone. The git ban on helpers is concurrency-scoped, so it lifts when the helper finds itself alone at wrap-up and the last agent out closes the door with the full wrap-up. The mirror case pauses too: a primary wrapping with live helpers stops at the commit and asks whether to sweep the helper's in-flight work, wait, or leave closing to the helper.
|
| |
|
|
|
|
| |
A second agent now discovers concurrency itself instead of being told: a stateless process scan (running agent processes, /proc cwd matched within the project root, own ancestry excluded) runs as the first action of every session, before any pull. Alone with no anchor is a fresh session, alone with an anchor is today's crash recovery, and not-alone skips startup and routes to helper-mode.org, the role-contract workflow. The scan also splits the previously ambiguous live-anchor signal into crashed versus concurrent primary.
Verified the signal live with four concurrent agents on this machine. The ai --helper launcher flag drops from mechanism to convenience. Known v1 limits recorded: sessions not running as local processes are invisible to the scan, and the match is process-cwd based.
|
| |
|
|
|
|
| |
Four loss windows the scoped-edit discipline doesn't cover: a primary file-wide hygiene pass silently clobbering a helper's concurrent edit (gate on live session-context.d/ files before any such pass), a new primary misreading helper dirt as leftover mess (surface live helper files at startup), crash recovery for shared-file edits (helpers journal each edit before applying it), and MEMORY.md's anchor-less read-modify-write index (memory writes stay primary-only).
Backstop: every file-wide pass snapshots to /tmp before modifying. lint-org and wrap-org-table already conform; todo-cleanup — the pass that moves whole subtrees — does not, and Phase 1.5 brings it up to the invariant.
|
| |
|
|
| |
The v0 draft covered identity and message targeting for concurrent agents but not spawn mechanics or write-safety for the shared files the session-context split doesn't isolate. I added a section for the motivating case (a second Claude in one project doing lookups and safe task updates): ai --helper spawn with automatic AI_AGENT_ID, a tiered read/write contract where helpers make scoped single-heading org edits and file-wide passes plus all git mutation stay primary-only, light helper startup, and helper wrap-up. Phase 1.5 sequences the slice independently of the runtime-neutral phases 2-6.
|
| |
|
|
|
|
|
|
| |
An XWayland client launched by --launch could send a configure request while the script tore down the headless output. Hyprland's damage path then dereferenced the removed monitor and the compositor aborted (Hyprland 0.55.2, coredump analysis in docs/design/).
The fix has two layers. --launch now forces the Wayland backend (DISPLAY unset, GDK and Qt steered to wayland) so no XWayland surface exists to race. Teardown also polls until the launched clients actually unmap before removing the output.
X11-only apps fail to map under the default, and some emacs builds are X11-only. The new --x11 flag allows XWayland for them, protected by the unmap wait. The no-window error hints at the flag.
|
| |
|
|
|
|
|
|
| |
From the template spec Craig wrote 2026-06-10 plus four refinements from his review of the first new-format prep. The doc is now exactly Heads-Up, Day's Priorities, and Meetings / Focus Blocks. Two run modes replace full-prep and standup-only: Create ends with a mandatory priorities review gate (disagreement there signals todo.org staleness), and Update refreshes a day when the world moves. Both run a triage-intake first when none ran in the last hour.
It retires the separate Standup Briefs and Upcoming Deadlines sections, the Anchor Tasks handoff, and the thin-link convention. Priorities entries now mirror their todo.org task heading and carry links and context in the body. Briefs nest under the standup they're reported in, with Blockers: None explicit. Meetings carry what to contribute and get, likely questions with answers, linked prep docs, and day-before prep blocks for unanswered questions. Focus blocks are linked menus, created the day before and marked free.
The spec and the decisions handoff land in docs/design/.
|
| |
|
|
| |
All five phases shipped today. The spec status flips to implemented with a history entry carrying the commit trail; the phase tasks become dated completion entries under the parent, which moves to DOING until the manual-testing checklist and the other machines' clone + timer setup land.
|
| |
|
|
|
|
| |
The KB leaves the ~/sync/org Syncthing share for its own git repo on cjennings.net (new decision D8). A systemd timer auto-syncs Craig's edits, agents pull before query and commit+push after write, machines replicate by clone (the work machine doesn't), and agent writes land under an agents/ subdirectory. Syncthing's no-history, no-gate, conflict-fork costs were the design's weakest accepted risks, and the phone constraint dissolved: mobile stays on on-demand doc drops to the ~/sync/phone share.
The amendment also folds in inclusion criteria plus a guided per-project memory sweep (Phase 1.5), a Success metrics section with a 30-day checkpoint, the seed node redefined as the KB's own documentation, and monthly hygiene automation (Phase 4). Phases renumbered 0-4. Implementation stays held pending the go-ahead.
|
| |
|
|
| |
Craig confirmed the denylist is complete at ~/projects/work alone (archangel is not work-scoped), which clears the spec's one remaining caveat. Phase 1 is unblocked, and implementation still awaits the explicit go.
|
| |
|
|
|
|
| |
I ratified all seven decisions: the org-roam KB is the shared agent substrate, the write boundary is read-shared write-scoped (work never writes), nodes are per-fact, agent writes land freely in the KB only, and harness memory stays as the ephemeral capture layer. The spec moves to docs/agent-knowledge-base-spec.org in spec-create format, superseding the 2026-06-05 draft.
A work-root denylist classifier routes writes: personal projects write, work and unknown projects refuse and report the redacted fact. Implementation is broken into three phases and waits on confirming the denylist contents.
|
| |
|
|
|
|
|
|
|
|
|
|
| |
artifact budgets
Two patterns kept failing in practice despite being documented (#40 praise asymmetry and the #38 terse cut), so I made the walk verifiable and closed the content gap behind tangled review text.
The high-recurrence set (#13, #37, #38, #40, #42) now gets per-pattern attestation receipts. The anti-AI audit runs after the terse pass so the audited text is the text that ships. Short personal-mode artifacts get a compact output format, and a write-back step puts the voiced text in the file the publish flow posts from.
Four patterns are new: #42 finding stems (one claim per sentence in review findings), #43 single-sentence paragraph cadence, #44 parenthetical asides, #45 declarative register marker. #37 exempts verdict formulas. #40 covers verification narration. #13 and #33 carry the self-discipline framing. A per-artifact budgets table makes terse a checkable budget instead of an adjective.
The profile gains paired entries with the approved worked examples, and commits.md plus no-approvals.org drop hardcoded pattern counts so the next addition doesn't re-drift them.
|
| |
|
|
|
|
| |
I folded the reusable, product-neutral checks from two emacs-d review passes into the canonical spec-review.org, so they survive the startup rsync and reach every project instead of living only in a downstream copy. The additions cover package-readiness and Makefile scope, actionable error strings, observability and diagnostics, long-running performance and failure-mode research, defcustom surface, a documentation plan, architecture weak-point mitigation, simplicity controls, extension/plugin developer experience, comparable-product sentiment, terminal-state discovery, CLI-wrapper value, and rollout/rollback, plus three reviewer principles and a generalizable-question harvesting rule.
The promotion is a pure superset. Every change adds or expands a generic check, nothing regresses. Project-specific findings stayed in the source spec. The handoff that asked for this is preserved under docs/design.
|
| |
|
|
| |
The spec adopts the existing ~/sync/org/roam/ KB (Syncthing-synced, 484 files) as the shared store agents read from and write to, so cross-machine memory sync comes for free instead of needing new infrastructure. It recommends the mechanics (queried as files, capture in harness memory then promote durable facts to the KB, a claude-rules pointer, an :agent: write schema) and leaves the work/personal write boundary for ratification. Supersedes the dedicated-repo and two-tier approaches for the storage-and-sync half.
|
| |
|
|
|
|
| |
On a flush resume, the SessionStart(clear) hook now reads .ai/notes.org key sections before the session-context anchor. The anchor carries session state. notes.org carries the project's standing knowledge (code-repo paths, conventions, key contacts) that a resumed session needs to act correctly. A resume with the anchor alone floundered on context notes.org already documents, hunting for a repo whose path notes.org records.
The hook guards on notes.org presence: when it's absent, the resume reads the anchor alone. flush/SKILL.md documents the same read order so the skill and the hook agree. The handoff rationale is preserved in docs/design/2026-06-02-flush-promotion.org.
|
| | |
|
| | |
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Walked the 14-item codex enhancement backlog with Craig and recorded
dispositions in docs/design/2026-05-28-rulesets-enhancement-backlog.org
under a new "Triage Dispositions" section. The triage TODO closed.
Accept (filed below as TODOs):
- Item #8 .aiignore for agent inventory exclusions ([#C] :chore:)
- Item #10 Workflow test harness for drift and integrity tests
([#C] :feature:)
Pilot or scope-limit (filed below):
- Item #5 Token-tier pilot on startup.org and triage-intake.org
([#C] :feature:)
- Item #7 Canonical/mirror drift detection via pre-commit hook or
make sync-check ([#C] :feature:quick:solo:). Rejected the wholesale
dedupe. The dual source is a feature.
- Item #12 make status only ([#C] :feature:quick:solo:). Rejected the
rest of #12.
Convention, not tracked:
- Item #11 script interface normalization (adopt as scripts are touched)
- Item #13 ADR layer (watch-and-wait on growth signal)
Reject with rationale (recorded in the backlog doc):
- Item #1 broader runtime-neutral arc. Heavy infrastructure for
marginal payoff at current use.
- Item #3 INDEX.org compression. Already token-efficient. Parallel
layer adds drift risk.
- Item #4 universal catalog.json. Looks valuable on paper, rots fast
in practice.
- Item #6 install manifest JSON. Current Makefile pattern works. Data
layer adds indirection.
- Item #9 project-facts.org. Duplicates notes.org's role.
- Item #14 local model profiles. Premature without a real use case.
Item #2 (per-agent live session files) was already filed earlier today
as the Codex Phase 1 [#B] TODO, slicing the broader item #1 arc.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Codex left a 14-item enhancement backlog for rulesets at
inbox/enhancements.org overnight. Moved it to
docs/design/2026-05-28-rulesets-enhancement-backlog.org as the canonical
reference and filed a single [#C] :spec: TODO to walk the items and
decide accept-as-TODO, fold, defer, or reject for each.
Two of the 14 already have homes: item #1 (runtime-neutral core) aligns
with the existing Generic agent runtime support spec (#16), and item #2
(per-agent live session files) is exactly the Codex Phase 1 [#B] TODO
filed earlier today. The triage focuses on the remaining 12.
Also cleared the last-session PROCESSED-prefixed inbox file
(open-tasks-hybrid-friction-cascade). The cascade+friction restructure
it tracked landed in this morning's open-tasks.org edits, so the
reference is no longer needed.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
All 16 open tasks bucketed and updated. 14 received autonomous Phase C
edits (13 type-tag additions per the new scheme plus #15's body refresh
for the accumulating pearl signal). Phase D adjudicated two priority
bumps and the morning's inbox spillover. Phase E stamped :LAST_AUDIT:
in notes.org Workflow State.
Autonomous Phase C:
- Tagged tasks 1-11 and 13-14 with their type tag (:feature:, :chore:,
:spec:), bumped LAST_REVIEWED to 2026-05-28. Tasks 12, 15, 16 already
carried type tags.
- Refreshed task #15 body to reference the four pearl pattern-catalog
notes now in docs/design/ (six worked patterns total).
Phase D adjudication:
- #15 (cross-project pattern catalog) bumped [#C] to [#B]. Pearl shipped
6 worked examples plus a synthesizing principle and is asking for
spec-review iterations. Design questions still open but evidence is
past the tipping point.
- Filed new [#B] :feature: TODO for the codex Phase 1 race-fix
(AI_AGENT_ID + session-context.d/<id>.org), lifted from the broader
runtime spec (#16). Phase 1 alone is low-risk and fixes a real
correctness issue under simultaneous agents.
Pearl intake (4 inbox files from this morning):
- Moved 0155 (patterns 4-5) and 0303 (pattern 6) into docs/design/
alongside the prior two pattern-catalog notes. Referenced from #15.
- Filed new [#C] :chore:quick:solo: TODO for pearl 0138's
--archive-done sweep at the start of open-tasks.org Phase A.
- Filed new [#C] :feature:solo: TODO for pearl 0226's spec-review.org
Phase 6 implementation-task enumeration.
- Deleted 0124. Already implemented this session as the audit-warranted
pre-step plus the LAST_AUDIT stamp now live in the workflow files.
|
| |
|
|
| |
Moved three inbox notes into docs/design/ so the task body links survive: pearl's two pattern-catalog handoffs and codex's v0 generic-agent-runtime spec. Added two corresponding TODOs under Rulesets Open Work, both [#C].
|
| |
|
|
|
|
| |
The spec recommended an Emacs keystroke mode (task-review.el). Implementation went the other way — a pure Claude workflow, no elisp — because the interactive mode would couple a rulesets-owned file to archsetup's init.el, and the daily Claude touchpoint already exists in daily-prep. I added a Revision section at the top recording the change: pure workflow, rulesets-owned, the task-review.org / open-tasks.org name swap, the staleness --list selection, and the startup nudge promoted to template-level. The elisp architecture and ERT sections stay as a record of the abandoned approach, flagged superseded.
The todo task moves to DOING with per-component status: everything but the smoke test is done, and component 3 (the elisp) is dropped.
|
| |
|
|
|
|
| |
docs/design/task-review.org captures the brainstorm output for a daily 5-min keystroke-driven review habit that walks 7 oldest-unreviewed top-level [#A]/[#B]/[#C] tasks per session, rotating through the list over ~12 days. Replaces wrap-it-up.org's date-coverage scan once implemented; the watchdog flips from "do all priorities have dates?" to "is the review habit happening?" with a 30-day threshold.
todo.org gets a [#A] entry at the top of Rulesets Open Work pointing at the spec, so the implementation work isn't lost. Six components in the spec's Next Steps: extract task-review-staleness.sh, replace the wrap-up section, author task-review.el in archsetup, author the workflow file plus INDEX entry, add the startup nudge, smoke test.
|
|
|
Four chained Claude Code skills covering the full architecture lifecycle:
arch-design Intake (stakeholders, scale, quality attributes, constraints)
→ candidate paradigms with honest trade-off analysis
→ recommendation + open-decision list
→ .architecture/brief.md
arch-decide ADR creation and management. Five template variants (MADR,
Nygard, Y-statement, lightweight, RFC). Lifecycle, review
process, adr-tools automation.
Forked from wshobson/agents (MIT). LICENSE preserved.
arch-document Full arc42-structured documentation (12 sections) from brief
+ ADRs + codebase. Dispatches to c4-analyze / c4-diagram for
Context, Building Block, Runtime, Deployment diagrams.
arch-evaluate Audits implementation against stated architecture.
Framework-agnostic checks (cyclic deps, stated-layer
violations, public API drift, forbidden deps) run on any
language. Opportunistically invokes language-specific linters
when configured (dependency-cruiser for TS, import-linter for
Python, go vet + depguard for Go). Never installs tooling.
Supporting docs at docs/architecture/:
- README.md suite overview, install steps, per-language linter install
commands (Python import-linter, TS dependency-cruiser, Go
golangci-lint, Java ArchUnit future, C/C++ IWYU future),
typical flow through the chain
- v2-todo.org deferred features (auto-gen linter configs, ArchUnit, CI
mode, DDD aggregate boundaries, visual dep graphs, retroactive
layering inference)
Makefile SKILLS extended with the four new entries; make install symlinks
them to ~/.claude/skills/ alongside existing skills.
Landscape: arch-decide fills the well-covered ADR bucket by adopting the
strongest community offering rather than reinventing. arch-design and
arch-evaluate fill gaps where no general-purpose skill existed. arch-document
fills the arc42 gap (C4 diagrams already covered by sibling skills).
|
