| Commit message (Collapse) | Author | Age | Files | Lines |
|---|
| |
|
|
|
|
| |
Phase 0 of the autonomous-batch (speedrun) build. todo-format.md now carries fixed cross-project definitions: :solo: is the autonomy/eligibility tag (buildable, agent-verifiable, no design deliberation — at most one or two quick upfront-answerable decisions, which the speedrun pre-flight Q&A batches), and :quick: is a ≤30-minute effort hint that never gates eligibility. task-review and task-audit now treat the tag assessment as mandatory — a pass that skips it is incomplete.
task-review's :solo: gate 3 also moves from "no upfront decision" to the no-deliberation form: the stricter wording predated the pre-flight Q&A decision and would have wrongly excluded tasks with a quick answerable question.
|
| |
|
|
|
|
|
|
| |
Rewrites every level-3+ DONE/CANCELLED/FAILED heading into a dated event-log entry from its CLOSED cookie, enforcing the todo-format depth rule that interactive closes and --archive-done (level-2 only) leave unapplied. A new lint-org checker (subtask-done-not-dated) flags stragglers, and the clean-todo, wrap-up, open-tasks, and task-review workflows now run the converter before archiving.
Removing the CLOSED cookie keeps a DEADLINE or SCHEDULED cookie that shares its planning line, rather than dropping the whole line.
From the .emacs.d handoff (2026-07-01 convert-subtasks bundle).
|
| |
|
|
|
|
|
| |
Verified ratio over tailscale: the roam clone and roam-sync timer are live and
syncing, matching velox. That closes the one-time per-machine setup, so the
standing open-instance note is cleared. The memory-sync task's only remaining
item is its manual-validation child, which needs a live cross-project session.
|
| |
|
|
|
|
| |
The rule claimed the agent can't reach the other daily driver, so the only move was to flag the drift. That's false: both machines are on the same tailnet, and this session repaired ratio's dotfiles over tailscale ssh. So the rule now says reach the other box directly to sync, verify, or repair, with flag-it kept as the fallback for when tailscale is down.
I added a section on the connection mechanics: tailscale status lists the nodes, the tailscale IP and MagicDNS name always resolve, a bare hostname only resolves when MagicDNS is configured locally, and the first connection needs StrictHostKeyChecking=accept-new.
|
| |
|
|
|
|
| |
A red test suite at the start of work poisons every later "did I break this?" check: you can't tell your own regressions from pre-existing noise, and the end-of-work green bar stops being readable. verification.md now asks for a clean suite run before work begins, not only before commit. start-work runs it as Pre-work step 0.3 against the reconciled base.
I added two carve-outs the original proposal lacked: a project with no suite has nothing to baseline, and a suite that can't run is the existing "When You Cannot Verify" case rather than a blocker. The step lands after the reconcile, so the baseline reflects the base the work is cut from. The Phase 4 TDD red is called out as expected, distinct from a baseline failure.
|
| |
|
|
| |
Any project with a codebase (including home and work) derives its codebase bugs' priority from a severity-by-frequency matrix. The priority is dictated, not argued. The mapping is fixed: P1 to [#A], P2 to [#B], P3 to [#C], P4 to [#D], with the bands defined per codebase. Features keep their roadmap judgment. Privacy, security, and safety defects are graded on severity alone. I preserved the worked-example bundle under docs/design.
|
| |
|
|
| |
.emacs.d resolves as emacsd and .dotfiles as dotfiles, in both inbox-send and the launch trigger. An exact basename match still wins, and --list shows the stripped name. triggers.md documents the same resolution so the spoken name is consistent across both.
|
| |
|
|
| |
I tightened the before-committing rule in verification.md and commits.md from "run tests" to "run the full suite as its own step and commit only on zero failures." A PreToolUse hook now backs it: it denies a test runner chained into git commit through an ungated connector, and lets the gated && form pass. I added a respond_deny helper. Tests cover the connector cases.
|
| |
|
|
| |
When a task needs Craig's location, run whereami on velox instead of asking. On any other host, fall back to asking. The rule covers the whereami-fails case (never fabricate a location) and keeps the geolocated address out of shared artifacts.
|
| |
|
|
| |
:BLOCKS: rulesets: was a malformed org tag, and the property form (:BLOCKED_BY: / :BLOCKS: carrying <project>: <what>) was more structure than the dependency needs. The blocking side now carries a plain :blocker: tag, mirroring :blocked: on the waiting side, with the which-project detail in the task body rather than a property. open-tasks.org reads the body for the blocking/requesting project; the scheme, the todo-format convention, and the inbox blocking-dependency handoff all move to the two-tag form. No property anywhere.
|
| |
|
|
|
|
|
|
| |
The :blocked: tag only marked the waiting side, so a blocker could stay unaware it was holding up another project: the dependency was visible to the one project that couldn't act on it and invisible to the one that could. This closes that gap. Setting :blocked: now requires a reciprocal inbox-send to the blocker, which files the work with a :BLOCKS: <project>: <what> property on its side. open-tasks.org surfaces :BLOCKS: tasks first, since clearing one unblocks another project (the highest-leverage pick), the mirror of pulling :blocked: tasks out of the cascade. Inbox process mode recognizes the blocking-dependency handoff shape, and the convention documents the resolution flow (drop :BLOCKS:, notify the waiter, who lifts :blocked:).
This works for any project pair, since the convention (todo-format.md) and the surfacing (open-tasks.org) live in the shared rule and workflow layer, not in one project.
Claude-Session: https://claude.ai/code/session_017PtX1nt1rtYVATuzmzBS4f
|
| |
|
|
|
|
| |
A task can be blocked by work owed by another project, and today it keeps getting recommended as "next" even though it can't move, while the blocker sits at low priority over there and the dependency stalls silently. Adds a :blocked: tag plus a :BLOCKED_BY: <project>: <what> property (todo-format.md) to mark the dependency, and teaches open-tasks.org Next Mode to pull :blocked: tasks out of the cascade pick and surface them in a dedicated "Blocked on other projects" section that names each blocker and offers an inbox-send nudge. Distinct from VERIFY, which waits on Craig rather than another project.
Claude-Session: https://claude.ai/code/session_017PtX1nt1rtYVATuzmzBS4f
|
| |
|
|
|
|
| |
Craig runs two daily drivers, ratio and velox, that need to stay in sync. The rule prompts every project session to consider whether an important machine-level change (dotfiles, a synced repo's clone or timer, a global config, a one-time bootstrap) needs to reach the other box, and to name the manual step when nothing carries it automatically. It carries the current open instance: the roam.git KB clone plus its roam-sync timer is confirmed on velox and still needs verifying on ratio.
Claude-Session: https://claude.ai/code/session_017PtX1nt1rtYVATuzmzBS4f
|
| |
|
|
|
|
|
|
|
|
|
|
| |
I merged process-inbox, monitor-inbox, and inbox-zero into one inbox.org engine. A shared core (value gate, skeptical review, disposition ladder, reply discipline, capture-guard, priority-scheme check) holds the logic that used to be duplicated and cross-referenced across the three files. Each mode (process, monitor, roam) references the core by name instead of restating it.
Every trigger phrase still works, now routing to a mode, so there's nothing to relearn. I added the interactive auto inbox zero mode: ask for an interval, run roam mode on /loop, acknowledge-only on an empty cycle, surface a find to a queue gated on a yes. The fully-unattended /schedule pass stays vNext, tracked separately.
I repointed every live caller (INDEX, protocols, startup Phase C, wrap-up Step 3, triage-intake, broadcast) at inbox.org and its modes, then deleted the three old files. triage-intake and no-approvals stay separate by design. The value gate, dispositions, capture-guard, and reply discipline all behave as before.
Built from the Ready spec. Workflow-integrity and sync-check pass on both the canonical and mirror trees, the stale-reference grep is clean, and the full suite is green.
Claude-Session: https://claude.ai/code/session_017PtX1nt1rtYVATuzmzBS4f
|
| |
|
|
|
|
| |
A non-Claude agent (Codex, a local model) reading the shared rules gets told it's "Claude" in prose that's really about whatever agent is running. I replaced the agent-as-actor references — "how Claude communicates," "when Claude needs the user to pick," a "wrong-context Claude session" — with "the agent."
Left untouched: file and path names (CLAUDE.md, .claude/, claude-rules/), and every reference in commits.md, which names Claude and Anthropic on purpose as the things never to attribute a commit to. Neutralizing those would invert the rule.
|
| |
|
|
|
|
|
|
|
|
| |
Editing the roam inbox on disk while Emacs has an indirect org-capture buffer cloned from it reverts the base buffer under the capture: the capture can't finalize with C-c C-c, and a freshly-typed item can be lost. inbox-zero Phase D edits that file, which Craig captures into constantly, so the collision recurs every session.
I added a capture-guard helper that asks the running daemon whether any CAPTURE buffer's base buffer visits a given file (file-equal-p, so symlinks and path spelling don't matter), exiting non-zero with the names when so. No reachable Emacs or no capture means exit 0, so it never blocks a write that was safe.
Phase D calls it before the pull, not only before the remove, because the ff-only pull also rewrites the file on disk and would wedge a capture the same way. On a collision an on-demand run stops and asks Craig to finalize or abort. The wrap-up sub-step skips the roam reconcile without blocking the wrap, since the items are already filed and the next run reclaims them.
emacs.md gains the inverse of the reload rule: don't yank a file out from under the daemon's live buffers.
|
| |
|
|
| |
The no-attribution rules covered AI credit but not incidental mentions of tooling filenames in commit prose. The case that bit: a .gitignore commit naming .claude, CLAUDE.md, and .ai in its message leaks the tooling layer into a public log. Adds a tooling-path-enumeration ban with that gitignore case named, extends the Before-Committing scan, and adds the missing paths to the protocols keep-out list. Both carry the file-is-the-change and private-single-user-repo exemptions, so a rule edit or a rulesets commit can still name what it touched.
|
| |
|
|
| |
The old rule dated a resolved VERIFY at every depth, including the top level. A level-2 dated header carries no keyword, so todo-cleanup's --archive-done can never archive it and task-review drops it from selection. Now a top-level VERIFY closes like any other top-level task (DONE/CANCELLED + CLOSED:), and dated rewrites are reserved for level 3 and deeper. Updated the rule and the three places that encoded the old behavior: todo-format.md, respond-to-cj-comments.md, and process-inbox.org. Also repaired two pre-existing level-2 dated headers.
|
| |
|
|
|
|
|
|
| |
A checklist step that is code the user runs (verification setup, bug repro, walkthrough wiring) now goes in an org src block instead of a list bullet, so the user executes it in place with C-c C-c and reads the result in the buffer rather than copy-pasting. Manual actions, prose context, and Expected lines stay as bullets between the blocks.
I scoped the Steps bullet in the same edit: it had said "one action per item" with the command as a bullet, which the new rule would contradict. It now names manual actions as the bullet case and points code steps at the src block.
From the smoke handoff 2026-06-12, worked out on its manual-verification walkthroughs.
|
| |
|
|
| |
A literal reading of cross-project.md could see its new propagation section (send synced-file edits to rulesets without being told) as conflicting with the file's stop-and-ask rule. One sentence reconciles them: ask-first governs work inside another project's scope, and an inbox drop is the sanctioned alternative to that, so it needs no confirmation.
|
| |
|
|
|
|
| |
A downstream edit to a rulesets-owned synced file (workflows, scripts, rules, protocols.org) is a stopgap the next template sync reverts. cross-project.md now documents the three-step propagation (edit locally, inbox-send the file to rulesets, include an intro note with the why and any companions to reconcile) so agents propagate a synced-file edit without being told.
From the .emacs.d handoff 2026-06-12.
|
| |
|
|
|
|
| |
Every project's task list now opens with a "[Projectname] Priority Scheme" section declaring both the [#A]-[#D] semantics and the tag vocabulary. The concept already lived in the tooling (task-audit enforces a declared tag set, process-inbox checks for the scheme before filing), but nothing required the section or fixed its name, so a list could leave [#A] and its tags undefined.
The set is declared, not fixed. A project adapts the priorities and tags to its own work. The floor is that both are spelled out.
|
| |
|
|
|
|
|
|
|
|
| |
Wide org tables overflow the page in exported PDF/docx, and hand-wrapping a cell into continuation rows is tedious and error-prone. The standard existed only as a work-project convention with nothing enforcing it.
claude-rules/org-tables.md carries the generalized standard: 120-column budget measured at render width (a link counts as its visible label and is never split), over-budget cells wrap onto continuation rows, and a rule sits under the header and every logical row.
wrap-org-table.el reflows a table to that shape mechanically. Columns shrink from natural width toward a floor of their widest atomic token, cells wrap link-safe, and rule-delimited continuation groups merge back into their logical row before re-wrapping, which makes the reflow idempotent. A table whose floors still exceed the budget reflows best-effort and stays flagged for restructuring.
lint-org.el gains an org-table-standard judgment check: width overruns and missing rules surface during the sweep with a pointer to the helper. Conformant wrapped tables don't false-flag, since the check reuses the helper's continuation-group reading. The check is judgment-only by design: reflowing is a visible layout change the sweep shouldn't make silently.
|
| |
|
|
|
|
|
|
| |
invariants
From the 2026-06-11 usage report. The debug skill gains a hypothesis-discipline contract (rank candidate causes by cheapest empirical test, run probes in parallel, report only confirmed findings) targeting the serial theory-cycling the report flagged. commits.md's pre-commit checklist gains a staged-files guard covering the untracked-set and canonical-vs-mirror conventions. A small tracked CLAUDE.md carries the rulesets mirror invariant at turn zero. Two [#C] pilots filed: a read-only morning ops orchestrator and a monthly session-harvest workflow.
The report's 500-token-cap finding was a mislabel: the underlying transcripts show 529 Overloaded and stream-idle-timeout errors with no token cap configured anywhere, so nothing to change there.
|
| |
|
|
| |
GFM renderers merge adjacent ordered lists and renumber them — a 1-3 content list followed by a 1-3 options list rendered as options 4-6, and the user's pick ("4") didn't exist in the author's numbering. Numerals now belong to the options list only, other enumerations in the same message use dashes or prose, and a prose lead-in sits directly above the options. An out-of-range pick is treated as a suspected render-merge, not an error.
|
| |
|
|
| |
Phase 1 of the agent KB spec. The rule carries the KB path and git discipline (pull before query, commit and push after write), the canonical rg commands, the one-node-per-fact write schema under agents/, the work-root denylist with the refusal contract, the inclusion criteria, and the capture-then-promote cadence. The existing Makefile RULES glob links it machine-wide; verified the link, a known-note query, and the conflict-file exclusion.
|
| |
|
|
|
|
|
|
|
|
|
|
| |
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.
|
| |
|
|
| |
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.
|
| |
|
|
|
|
|
|
|
|
| |
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.
|
| |
|
|
|
|
|
|
| |
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.
|
| |
|
|
| |
A freshly-added authinfo.gpg entry still reads as missing in the running daemon until auth-source's result cache expires. It caches both positive and negative lookups for auth-source-cache-expiry (default 7200s), so a key you just synced in keeps erroring as "not set" with no sign the file is fine. Clear it with (auth-source-forget-all-cached) and re-trigger the lookup. I hit this diagnosing a Linear/pearl API-key error after the key landed via a dotfiles sync.
|
| |
|
|
| |
Numbered choice lists put the recommended option at item 1 so the common case collapses to one keystroke. Skip when the question is free-form, when a directive's already been issued, or when no option is clearly better. Say so plainly in that case.
|
| |
|
|
|
|
| |
The no-popup-menus rule in interaction.md was too easy to forget, so the popup kept slipping back into choice prompts. I added a PreToolUse hook on AskUserQuestion that denies the call outright and returns the rule as the reason, which routes choices back to inline numbered lists. Since ~/.claude/settings.json symlinks to this repo's .claude/settings.json, the hook is machine-wide and version-controlled across machines.
I documented it under the rule in interaction.md, including the consequence: the deny is unconditional, so the old "use the popup for this one" exception now needs the hook disabled via /hooks first.
|
| |
|
|
|
|
|
|
|
|
|
|
| |
fragments, formatting
The personal voice patterns only ran for commits and PRs, so the emails and documents I author never got my actual writing voice. General mode deliberately skips them. I added a third mode, prose, that applies my voice patterns to prose I write or send without dragging in the publish-artifact mechanics that misfire on free text.
The modes now nest. General (#1-31) handles anyone's prose, prose adds my voice patterns (em-dash zero-tolerance, contractions, semicolons to periods, sentence-split, felt-experience cut, fragment rewrite, terse-cut, no-emphasis-formatting), and personal adds the three artifact-mechanics patterns on top (first-person rewrite, public-artifact scope flag, praise/correction asymmetry). Those three stay personal-only because they assume a commit or PR: a document is legitimately third-person, a journal has no public-scope concern, and praise/correction asymmetry is a PR-review rule.
Three gaps closed along the way. #13 (em-dash) was "use fewer". It's now zero-tolerance in prose and personal modes, and the rule holds inside examples and quoted text, not just running prose. #37 (every prose sentence needs a subject and a verb) was locked to personal mode. It now applies to my prose too. And #41 is new: I make points with words, not bold or italics or underscores, so emphasis markup gets rephrased so the stress lives in the wording.
I updated commits.md to match. The publish flow still uses personal mode, but the pattern count is now 41 and the personal-only set is the three artifact-mechanics patterns.
|
| |
|
|
|
|
| |
Craig's terminal renders Markdown bold and inline-code spans as reverse video, which is hard to read. I added a rule to interaction.md: in conversational output, write command names, paths, and key chords as plain text, and lean on headers, dashes, parentheses, and quotes for structure. It governs chat output only, not the Markdown source of the rule and spec files he reads in an editor.
I also made the keybinding-display example plain text so the convention shows the format the way it should appear in chat, with a pointer to the new rule.
|
| |
|
|
| |
I added a rule for how to present a keymap's bindings when asked to show them. The format is a bulleted list grouped by prefix level: a General header at the top that lists the sub-prefixes leading into each category, then one section per category. Every bullet carries the full chord, the bound command, and the which-key label, so the written view matches what which-key shows on screen.
|
| |
|
|
| |
When a verification gap needs the user's hands or eyes — interactive UI a script can't drive, a live service, visual rendering — describing the steps in prose isn't enough. I added a section to verification.md that says to write them as a "Manual testing and validation" task: one sub-header per test, each with a descriptive title, what we're verifying, the steps as a bullet list, and the expected result. If a test fails, the user writes the actual behavior, flips the header to a TODO, and promotes it, so a failed check becomes a tracked bug in one step.
|
| |
|
|
| |
Adds claude-rules/emacs.md documenting how to push a module edit into the long-running emacs daemon via emacsclient instead of restarting and re-opening files. Covers the reliable case (function redefinition), the caveats that bite (defvar defaults don't re-apply, use-package :config re-runs, faces need re-applying, baked or rendered state like the dashboard buffer must be regenerated), and the reload-and-verify loop. Captured after a long dashboard session where a stale buffer repeatedly masked working code.
|
| |
|
|
| |
The publish flow made the /voice pass mandatory but had no fallback for when the skill isn't installed, which a fresh or partial environment can hit. Added one to the Single-skill gate: when /voice can't run, walk the patterns inline (the flow already names which matter), state the skill was unavailable and the pass was applied by hand, and flag the missing skill for install. The gate is the pattern walk, not the tooling — the skill is just the convenient way to run it. The original audit framed this as a "humanizer unavailable" gap. Humanizer is /voice now, so only the /voice-unavailable case remained.
|
| |
|
|
|
|
|
|
|
|
|
|
| |
The global commits.md carried DeepSat-specific publishing steps — Linear ticket-state moves, the Slack notification protocol with its channel ID and engineer names, the deepsat.ghe.com host, the team merge norm. Those are symlinked into every project on the machine, so they sat as dead weight in personal repos and risked misfiring where there's no Linear ticket to move or Slack mpdm to ping.
I split them out. commits.md keeps the universal skeleton (identity, attribution, commit format, the review-and-publish gate, verification) and replaces the team steps with seams: "run the project's publishing overlay here if it defines one," the same pattern startup.org uses for startup-extras. A project with no overlay runs the complete flow, just without ticket and chat integration.
The DeepSat specifics move to teams/deepsat/claude/rules/publishing.md. That file is not a global rule — install-team.sh copies it into one project's .claude/rules/ (make install-team TEAM=deepsat PROJECT=...), keyed on the PROJECT argument, so only the named project gets it. Location decides distribution: claude-rules/ is the global-symlink set, teams/ is targeted-copy, so the overlay reaches DeepSat and nowhere else.
The startup freshness check (sync-language-bundle.sh) now covers team overlays alongside language bundles: a process_bundle function handles both, with a team syncing only its own rule (no generic rules, hooks, or settings — those belong to a language bundle). A drifted overlay rule auto-fixes from canonical at the project's next startup, the same mechanism language bundles already ride.
Tested: 3 new bats cases (team overlay clean / drifted-and-fixed / does-not-pull-generic-rules) on top of the 11 existing; install-team + sync verified end-to-end against a temp project. make test green, shellcheck clean.
|
| |
|
|
| |
subagents.md assumed the Agent tool exists. A new "Pre-Dispatch Checks" section adds two gates before any spawn: Availability (no Agent capability means do the work in the main thread under the same scope and constraints the contract would enforce) and Cost (when writing the full contract costs more than the task, do it inline). Both cross-reference the existing "Don't Subagent At All" guidance rather than duplicating it.
|
| |
|
|
| |
Two additions. An "Escalation Beyond Category and Pairwise" section adds property-based testing (for invariants across a broad input domain) and mutation testing (for when high coverage hides thin assertions), both as escalation paths rather than always-on gates. And the "I need to spike first" excuse is formalized into a disciplined spike protocol: TDD stays the default, but a spike is sanctioned only when timeboxed, not committed, and followed by the first failing test before productionizing.
|
| |
|
|
| |
verification.md required running tests/lint/typecheck/build before claiming done, but said nothing about what to do when a command can't run. A new "When You Cannot Verify" section requires a four-part report (command attempted, why it couldn't run, risk left unverified, and the smallest next command for the user) and states the principle that a check that didn't run is never reported as a pass.
|
| |
|
|
|
|
| |
The .ai/-tracking check used to decide two things at once: which voice patterns ran and whether the approval gate fired. In a team repo that meant losing the 8 personal patterns and the gate together. I split them. Publish artifacts (commit messages, PR titles and bodies, PR review comments) always run /voice personal now, because they go out under my name regardless of the repo. The .ai/ check decides only the gate: applied in my personal repos, skipped for velocity in shared ones.
The gap this closes: a team-repo PR comment used to skip pattern #39, the public-artifact scope flag, which is exactly the check that matters most when teammates can read it.
|
| |
|
|
|
|
| |
review-code was a command with disable-model-invocation set, so the model could never reach for it on its own. Every time a review fit the moment, the agent had to hand back to me to type the slash command. Moving it to a skill makes it model-invocable while it stays slash-invocable as /review-code.
git mv keeps the file history (99% rename). The frontmatter drops disable-model-invocation and gains name: review-code; the body is unchanged. It also drops the discovery-check paragraph in commits.md, which only existed to find the command on disk when it was missing from the skills list, moot now that the skill shows up there.
|
| |
|
|
|
|
|
|
| |
Adds claude-rules/triggers.md as the home for phrases the user says from any cwd to invoke a cross-project action. First entry: "launch project X" → run the ai script in single-project mode targeting the matched basename. Ambiguity handling: list candidates and ask rather than guess.
The trigger phrases already in protocols.org ("Let's run the [X] workflow", "Wrap it up") are project-scoped. They assume an active .ai/ session. Cross-project launchers don't fit that layer; they need to work from any cwd, including outside any project. None of the existing claude-rules files (commits.md, testing.md, verification.md, subagents.md, interaction.md, cross-project.md, todo-format.md) had a clean fit either. A new file is the smallest architectural change.
Makefile picks up new claude-rules/*.md files via wildcard, so no Makefile change needed. make install created the symlink at ~/.claude/rules/triggers.md; make doctor reports 39/0/0.
|
| |
|
|
|
|
| |
I added claude-rules/working-files.md as the canonical convention. In-progress task artifacts live in working/<task-slug>/ under the project root. On task completion the files get renamed individually with a YYYY-MM-DD-<slug>-<descriptor> shape and moved flat into assets/ (or the area-specific assets/). Never rename the directory as a substitute for filing, since that loses the flat-searchable property of assets/.
claude-templates/.ai/protocols.org gained a short reference to the rule so it propagates to every project's .ai/protocols.org on the next startup rsync.
|
| |
|
|
|
|
|
|
| |
Step 1 told the agent to run /review-code but didn't say what to do when the skill exists on disk yet isn't in the session's available-skills list. The list covers plugin-installed skills only. User commands under ~/.claude/commands/ are routable as slash-commands but don't appear in it, so the agent could declare /review-code unavailable and fall through to the trivial-one-liner exception in Step 2.
The new Discovery check tells the agent to verify both ~/.claude/commands/review-code.md and ./.claude/commands/review-code.md on disk before declaring the skill unavailable, and surface the mismatch rather than auto-skipping.
Also drops three absorbed or stale inbox files: the skill-discovery handoff (signal absorbed by this edit), the missing-inbox-dir handoff (already resolved by 470085f), and a stale date-coverage scan output (deferred until the task-review habit lands).
|
| |
|
|
| |
The MCP's `conversations_add_message` tool takes the raw mrkdwn body in `payload`, not a Slack API JSON envelope. Wrapping the body in `{"text": "..."}` posts the literal JSON to the channel — `<@>` mention strips, escaped `\n` renders as a literal `n`, and the `text:` key prefix shows up in the post. Step 8 now documents the format + mrkdwn list so PR-review notifications don't garble.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Todo-format.md gets a "Completion — depth-based" section that
codifies what was implicit before: top-level (* and **) DONE tasks
stay task-shaped (DONE + CLOSED line), sub-tasks (*** and deeper)
flip to dated event-log entries, and VERIFY is the documented
exception (always dated-rewrite regardless of depth). The section
includes worked examples and the rationale for depth-based over
keyword-based.
The cj-comments skill's completion section now references the
canonical rule instead of restating the depth-based rules inline.
The three-bullet recap in the skill stays as a quick scan, but the
details and examples live in one place.
|
