| Commit message (Collapse) | Author | Age | Files | Lines |
|---|
| |
|
|
| |
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.
|
| |
|
|
|
|
| |
The 2026-06-10 sweep shipped without Signal: a standalone signal-cli receive hung on the account lock while the signel daemon owned it, and the failure looked identical to a quiet source. The engine now renders any failed, hung, or skipped scan in a SCAN FAILED banner at the top of the summary. Quiet means the scan ran and found nothing.
The signal plugin now detects which path owns the account before scanning: when the signel daemon is live it queries chat buffers through Emacs, and the standalone draining receive runs foreground-only when it isn't. The telegram plugin gets an at-a-glance lifecycle (docker-mode launch, scan, send, shutdown only if the scan started the server) and treats a real DM from a work contact as Action.
|
| |
|
|
|
|
| |
A Day's Priorities entry now carries the URL the task is done through (payment portal, doc, PR, form), not just the thin link to its todo.org home. When that URL is missing from the todo.org body, the prep build finds it and adds it to both. Meeting lines in Meetings / Work Blocks get an org link to the join URL from the calendar event's conference data, so joining is one click from the prep doc.
Both rules came out of working the 2026-06-10 prep doc, where paying an invoice meant a Gmail dig for the payment URL first.
|
| |
|
|
|
|
| |
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.
|
| |
|
|
| |
Meeting-prep promotion session. The completed task moved from Open Work to Resolved in the todo cleanup.
|
| |
|
|
|
|
| |
Meeting-prep was project-only. I generalized its project-specific references (transcript-home path, issue tracker, knowledge file, worked-example doc) to neutral terms and moved it into claude-templates so any project's .ai/ picks it up. The pre-wire supporting doc travels beside it as meeting-prep.pre-wire.org.
I added the workflow entry and trigger phrases to INDEX.org. I flipped daily-prep's two conditional meeting-prep references, and its trailing changelog claim, to direct links now that the workflow ships as a template.
|
| | |
|
| |
|
|
| |
tag meeting-prep solo
|
| |
|
|
| |
A green check only covers the files the gate actually ran on. When a lint, test, or format gate uses a hand-maintained file list instead of a glob, a newly-added file is silently skipped and the gate still reports clean. Added a subsection naming the failure mode (a Makefile path list, a pre-commit files: regex, a CI matrix, a coverage include list) and the check it calls for: when you add a file, confirm each gate discovers it before trusting the green.
|
| |
|
|
|
|
| |
Craig clarified what :solo: means. The old third gate ("the outcome is verifiable locally, no ... confirmation that the result is right") read literally disqualified every task, since Craig spot-checks everything regardless of the tag. It conflated "Craig will also check" with "only Craig can check."
The three gates are now buildable, verifiable by Claude, and no upfront decision. The fix is decoupling Craig's routine spot-check from the determination: a task Claude builds and verifies itself, leaving a manual-testing reminder for the residual human-in-the-loop confirmation, is solo. The disqualifier is having no verification path of Claude's own, a result only judgeable by Craig's eyes. task-audit.org Phase C already defers here for the definition, so this is the one edit site.
|
| |
|
|
|
|
|
|
| |
A gitignore-mode project only ignored .ai/. CLAUDE.md was left untracked but not ignored, so an accidental git add or a codify run could still commit a personal CLAUDE.md, the private rule copies under .claude/, or an AGENTS.md. install-ai now ignores the whole set (.ai/, .claude/, CLAUDE.md, AGENTS.md) at bootstrap, line-idempotent so an existing .gitignore isn't duplicated.
.claude/ goes in the set because it's rulesets-owned (copies of claude-rules/*.md plus the language bundle's rules, hooks, and settings), re-synced from rulesets every startup, so git isn't how it travels. Ignoring it also keeps those private rule copies out of the repo, which ignoring CLAUDE.md alone would miss. The gate is unchanged: track-mode projects (personal/doc repos, team repos sharing config) keep tracking the set.
sweep-gitignore-tooling.sh backfills the set across existing gitignore-mode projects, idempotent and skipping track-mode by design. It warns when a now-ignored path is already tracked, since the ignore won't untrack it. protocols.org states the policy once.
|
| |
|
|
| |
Promoting meeting-prep to a template needs its project-specific references generalized first, so it's a follow-up rather than part of the daily-prep additions.
|
| |
|
|
| |
I folded in three additions from the Manager Tools prep research. Phase A now fetches the prep day plus the next 5 days, and a new Phase 1 sub-section scans that window for meetings that need prep, scheduling traps, and focus blocks to protect. It's a scan-and-flag pass that feeds Heads-up and Anchor Tasks, not a new prep-doc section. Phase 3 sub-step 3a pulls one important-but-not-urgent task per day for a 15-minute chunk, so strategic work lands in small daily pieces instead of getting displaced by the urgent. Phase 1 item 5 reframes "attend" as "contribute" and adds a send-regrets gate for meetings with no objective. The look-ahead's meeting-prep references stay conditional ("where the project has one") since meeting-prep is project-only for now.
|
| |
|
|
| |
A workflow that outgrows its inline content stores the overflow in a sibling <workflow>.<suffix>.org, linked inline. This reuses the engine.plugin dot-glob, so the drift-check and discovery glob already treat the file as owned rather than an orphan. It also travels with the workflow on promotion. I extended the INDEX drift-check note to name supporting docs alongside source plugins.
|
| |
|
|
| |
A generated config gets read in a different namespace than the one that wrote it. Build host, chroot, image runtime, target system, container, and VM are distinct, so a path valid where the config is generated can be absent where it's read. I added the check to the Architecture & maintainability dimension.
|
| |
|
|
| |
Two work-project handoffs filed as backlog tasks: a SessionStart-on-clear auto-resume hook (after /flush), and wide org-table handling via an auto-wrap helper, a width lint, or a tightened standard. Inbox-process marker updated.
|
| |
|
|
| |
cross-project-broadcast handled tooling and rule announcements but had no shape for the situational case: a life or work event I want every project's agent to know, said once so none is missing context when I next talk to them. I renamed it to broadcast (helper and test alongside) and split it into two modes over the same fan-out plumbing. Announcement keeps the rigid capability template. Situational carries a general-not-comprehensive summary plus a fixed receiving-agent contract: record it in notes.org, hold it time-boxed or standing, apply on the project's own judgment, ask follow-ups at startup. The broadcasting agent does no per-project relevance analysis. Each receiving agent decides what the event means for its own work.
|
| |
|
|
| |
spec-create is the front of the spec-create → spec-review → spec-response trio: the author writes a short design spec before non-trivial code, shaped to pass spec-review's readiness gate. It runs a when-to-spec proportionality gate first, then problem-first framing, design with forced alternatives and inline mini-ADR decisions, implementation phases with acceptance criteria and a readiness-dimensions menu, a terseness pass, and a hand-off self-check against the review rubric.
|
| |
|
|
| |
The real elisp packages drive eask in their Makefiles, but the bundle listed eldev as a build tool and left eask out, pointing a fresh session at the wrong tool.
|
| |
|
|
|
|
| |
The skill ran one flat criteria set on every change, so a concurrent, DB-backed service got the same review as a shell script, and the checks that matter for the service (concurrency, performance, auth, API compatibility) weren't in the set at all.
Phase 3.5 splits the checks into a universal core that always runs plus nine modules that activate on the project's review profile. A project declares the profile with a "Review profile:" line in its CLAUDE.md. Absent that, the skill auto-detects from framework, ORM, concurrency, and manifest signals and recommends declaring one. The security and dependency modules defer their depth to /security-check.
|
| |
|
|
|
|
|
|
| |
I added a Telegram source plugin so the triage-intake sweep covers Telegram alongside Signal, cmail, Gmail, calendar, and PRs. Telegram is personal messaging, so it's a general plugin that syncs to every project.
Unlike signal-cli, Telegram has no headless CLI here, so the plugin drives telega.el inside the running Emacs daemon over emacsclient. It records whether telega was already live and shuts it down only if the scan started it, leaving an active session alone. Two sharp edges are documented in the plugin: the tdlib server can SIGSEGV on the initial sync, where docker mode is the fix, and the scan reads the cached telega--chats hash so a dead server still reports unread state instead of going blank.
I also added Telegram to the engine's general-plugin list.
|
| |
|
|
|
|
|
|
| |
I added a Signal source plugin so the triage-intake sweep covers Signal alongside cmail, Gmail, calendar, and PRs. Signal is personal messaging, so it's a general plugin that syncs to every project.
It needs no wrapper script, unlike cmail. signal-cli is already a full CLI, so the plugin drives receive and send directly. The scan filters signal-cli's JSON down to real messages and drops the sync, receipt, and typing noise. One sharp edge is documented in the plugin: signal-cli receive drains the server queue, so the triage gets one shot per message. Signal Desktop and the phone keep their own copies, so nothing's lost.
I also added Signal to the engine's general-plugin list.
|
| |
|
|
| |
validate-el.sh dumped the full ERT batch output to the terminal on a red test, every backtrace frame, flooding the pane. It now prints a short summary there: the run tally plus the failing test names and their file:line. Claude still gets the full backtrace through additionalContext. fail_json takes an optional fourth argument for the terminal echo. Paren and byte-compile failures stay short, so they still print in full.
|
| | |
|
| |
|
|
|
|
| |
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.
|
| |
|
|
|
|
| |
I added Phase 6 to spec-response: once the author confirms a spec is Ready, file the full implementation-task breakdown in todo.org rather than leave a Ready spec nobody can act on. The phase creates one task per implementation phase, runs a completeness pass against every acceptance criterion and principle rule, marks :solo: only where the agent can build and verify end to end, collects the rest under a Manual-testing task, and defers outward-facing publish steps until the user confirms.
This is the author-side complement to spec-review's Phase 6, which emits the drop-in task block. Review produces the block, response files the work.
|
| |
|
|
| |
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.
|
| |
|
|
|
|
|
|
|
|
| |
A good interaction-design pattern surfaces in one project, then gets lost, so the next project re-derives it. Pearl shipped six worked examples and the principle they share, but they sat as four raw notes in docs/design/ with nothing making them reusable.
I added a patterns/ directory, one file per pattern, each with frontmatter (name, principle, problem, tags, source, examples) and a Problem / Do / Anti-pattern / Applicability / Related body. The six seeds: one-prompt-picker-typed-prefix, transient-state-buttons, no-empty-input-as-meaningful, label-matches-behavior, default-most-common-friction-proportional, collapse-orthogonal-prompts. Patterns 3-6 cross-link as facets of one root principle: the choices the user has should all be on screen, accurately labeled, ordered by what they'll most often want, with friction sized to the cost of being wrong.
The catalog lives in rulesets because every project's agent already pulls from here. A thin claude-rules/patterns.md pointer tells the agent the catalog exists and when to consult it, so it reads one pattern on demand instead of carrying all of them. The pointer auto-installs via the Makefile RULES glob. README.org holds the root principle, the frontmatter contract, and the capture-on-landing/promote-on-review intake cadence.
The patterns are .org files with #+KEYWORD frontmatter. The pointer stays .md because the rules layer and the Makefile glob require it. Seed patterns keep their concrete Elisp examples. Generalize on the second caller, not up front.
|
| |
|
|
|
|
|
|
| |
A skill added to rulesets and pushed reached each machine's files on the next pull, but not its ~/.claude symlink. make install only links what isn't already linked, and a git pull doesn't run it. So a new skill stayed silently uninstalled until someone re-ran make install by hand. The flush skill sat in that gap from 2026-06-02 until a manual install today.
I added a make install step right after the Phase A.0 rulesets pull, the step that brings the skill's files in. It's idempotent: skips already-linked targets, links only what's new, and only writes symlinks under ~/.claude and ~/.local/bin. A grep keeps the all-skip case to one quiet line. Link, relink, and WARN lines get surfaced, and a new Phase C bullet handles them: a skill the harness already picked up mid-session is noted as available, one it didn't gets a restart prompt, a non-symlink collision goes to the user.
Now "add a skill, commit, push" is enough to reach every machine on the next session. The step also covers a newly-added rule or script, since make install links those too. The canonical template and the .ai/ mirror both carry the change.
|
| |
|
|
|
|
|
|
| |
Pattern #38 (omit needless words) was one of 41 walked mid-list, so it got glossed as a checkbox instead of the real "cut it in half" sweep. A commit message went out today needing two manual Orwell-walk requests before it read terse, even though /voice personal had run.
The rule already said to cut hard. The gap was position. Buried in the middle of the walk, #38 lost to the categorical detectors that come back clean while the text still runs a third too long.
I moved it to an explicit standalone last step in the SKILL.md Process, run after every other pattern and right before the draft is shown. Now the first draft the user sees is already terse, not terse only after they ask. Profile §38 gains an execution-position rule and a history entry. The commits.md publish gate points at #38 as the last pass.
|
| | |
|
| |
|
|
|
|
| |
The startup bundle sync re-dropped from-rulesets-coverage-makefile.txt into a project's inbox on every run, even after the project had adopted the targets. inbox_drop only treated the fragment as adopted if coverage-makefile.txt still sat at the project root or waited in the inbox. But install-lang tells users the opposite: copy the targets into your Makefile, then delete the fragment. So a project that followed the documented path got the drop re-suggested forever (deleted three sessions running in one case).
I guarded the drop so a project Makefile that already defines the distinctive coverage-summary target counts as adopted. The check lives at the call site, keeping inbox_drop generic. Added two bats cases: targets-in-Makefile suppresses the drop, an unrelated Makefile still gets it.
|
| |
|
|
| |
Both descriptions ran past Claude Code's per-entry listing cap (~1536 chars), so the menu blurb got truncated. start-work was 1680, respond-to-cj-comments 1568. I trimmed them to 1513 and 1518, compressing only descriptive prose. Every "Use when" and "Do NOT use" routing trigger stays verbatim, so triggering is unchanged.
|
| |
|
|
| |
Promoted the flush workflow into rulesets as a /flush skill plus a canonicalized SessionStart(clear) hook, then ran the four open :solo: tasks: start-work Justify and Approach gate additions, a task-review chain on task-audit, and lint-org follow-ups reconcile-on-write.
|
| |
|
|
|
|
|
|
|
|
| |
Every run appended a fresh dated "lint-org follow-ups" section with line-number-keyed entries, so the follow-ups file grew an unbounded pile of near-duplicate sections, kept entries whose finding had since resolved, and broke whenever the target file's line numbers shifted. Running an audit against a large todo.org surfaced exactly that drift: dead-link flags pointing at docs that now exist, and three stacked dated runs for one file.
Now lint-org rewrites the current file's section from the current run. Findings that no longer reproduce simply are not re-emitted, re-runs dedupe to one section, and entries key on checker plus message with the line as a trailing annotation, so a finding survives line shifts as the same entry. Other files' sections are left intact, and the strip step tolerates the old dated-header shape so existing follow-ups files migrate on first run. This changes the follow-ups file from an append-only log to the current outstanding findings per file.
task-audit's Phase C link-hygiene step now also reaps a matching dead-link entry when it fixes or verifies the link, scoped strictly to dead-link entries, so the audit and the follow-ups file stop drifting between lint runs.
Five follow-ups tests cover record-by-content, dedupe across runs, drop-on-resolve, and preserve-other-files. Mirrors synced.
|
| |
|
|
| |
A task audit verified the surviving tasks are factually honest but left their relevance and priority untouched, so keeping the list lean still needed a separate task-review run. Added Phase F: after the audit stamps :LAST_AUDIT:, run task-review on the oldest-unreviewed batch in the same pass. The two stay distinct (audit owns facts and :LAST_AUDIT:, review owns relevance and the per-task :LAST_REVIEWED:); chaining just refreshes both markers in one invocation. open-tasks does not invoke task-review, so nothing there needed to change. Mirror synced.
|
| |
|
|
| |
The Approach gate planned tests, migration, and commits but never asked whether the work needs a design spec it does not yet have. Added item 5: a spec is warranted for large or wide-surface work, unresolved design questions, or a new interface others build on. For a big task it is never a silent skip — the approach summary must state why no spec is needed, so the call is visible and challengeable at the gate. Small contained tasks pass without comment.
|
| |
|
|
| |
The Justify gate covered Downsides and Alternatives but had no forced verdict on whether the work should happen at all. Added item 9: surface the top three objections when real ones exist, or say so in one line when none rise to a genuine objection, rather than manufacturing three. Building the case against the work is cheapest at this gate.
|
| | |
|
| |
|
|
|
|
| |
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.
|
| | |
|
| | |
|
| | |
|
| |
|
|
|
|
|
|
|
|
| |
Flush is the checkpoint half of the wrap/restart rhythm. It refreshes the session-context anchor in place, the user runs /clear, and the session resumes from the anchor instead of starting cold. One logical session stays alive across a /clear boundary without the archive-and-commit of wrap-it-up or the full cold boot of startup, which buys cheaper tokens and a sharper context window.
The mechanism splits into two halves around /clear, which wipes the conversation so nothing runs straight through it. The /flush skill is the pre-clear half: dump live state, refresh the anchor's Summary, append a dated flush marker, verify the write landed, then prompt the user to /clear. The agent can initiate at a clean task boundary on its own judgment, but /clear is user-only, so the agent does the work and the user supplies the single keystroke. The session-clear-resume.sh hook is the post-clear half, a SessionStart matcher=clear hook that points the fresh session at the anchor to resume, or at startup when no anchor exists.
I packaged the pre-clear half as a skill rather than a project-workflow doc so both halves are global. The hook was already global, so /flush is now callable by name from any project with no per-project sync.
The hook is canonicalized under hooks/ and symlinked into ~/.claude/hooks/, matching precompact-priorities.sh. settings.json wires the SessionStart entry, and settings-snippet.json carries it so a fresh machine wires the hook on make install-hooks.
|
| | |
|
| | |
|
| | |
|
| |
|
|
|
|
|
|
|
|
|
|
|
| |
The Go bundle was coverage-slice-only. Because it shipped no rule files, sync-language-bundle.sh (which fingerprints a project's bundle by spotting one of its rule files in .claude/rules/) couldn't detect it, so the coverage slice it did ship never stayed in sync. Adding the rules is what makes the bundle sync-maintainable, which was the point.
Brought Go to the full tier, matching elisp:
- claude/rules/go.md and go-testing.md, the style and testing rules (table-driven tests, go test -race, errors.Is over message matching, how the coverage slice fits). These two are also the sync fingerprint.
- claude/hooks/validate-go.sh, a PostToolUse hook that runs gofmt and go vet on each edited .go file. go vet type-checks, so compile and syntax errors surface at edit time. It deliberately doesn't auto-run tests, since a package's tests can be slow or integration-tagged and shouldn't fire on every keystroke.
- claude/settings.json, Go permissions plus the hook wiring.
- githooks/pre-commit, a secret scan and a gofmt check on staged .go.
- CLAUDE.md, the seed.
validate-go.sh is TDD'd by scripts/tests/validate-go.bats: a clean file passes, gofmt and vet failures both block with the JSON payload, and non-go, missing, or empty paths are ignored. I updated install-lang.bats test 7, which asserted Go installs no CLAUDE.md, to check the full bundle instead. Verified with a real install into a throwaway project and a green make test.
|
