crisradu75/cla

Open a Markdown, text or HTML document, or a whole OpenSpec change, as a local page where the user selects passages and comments; the comments are read back and worked through. Never edits the source. Triggers: /cla:annotate, 'let me annotate this', 'let me review change X as a block', 'read my annotations'. Not a pre-implementation review (review-change).

Compact the session into a resumable briefing in cla.io/checkpoints/: decisions and why, what landed, what is in flight with its next action, open threads. Points at commits and PRs; never copies secrets. Triggers: /cla:checkpoint, 'checkpoint this session', 'I am running out of context'.

Set up or refresh a repo for the cla plugin: create whatever is missing of cla.io/ without overwriting anything there, seed OpenSpec authoring rules, write the economy rules file .claude/rules/cla-economy.md (replaced whole on every run), and populate or reconcile cla.io/project-facts.md from the repo's own manifests. Changes any other existing file only on the user's yes, apart from appending one line to .gitignore. Run with /cla:cla-setup.

Review the whole current session for failures, check each against the rules earlier runs wrote, propose up to three fixes (tools and hooks before docs) to apply with one prompt, and log the fixes and re-offenses to cla.io/. Run with /cla:codify-learnings.

Find a failure's root cause instead of guessing: a deterministic repro loop, ranked falsifiable hypotheses, one-variable instrumentation, the regression test before the fix. Also reached when a Test phase stalls on one cause. Triggers: /cla:diagnose, 'why is this failing', 'find the root cause'.

Capture the user's notes one at a time, ground each read-only in the repo, and consolidate them into a dated triage doc under cla.io/feedback/ grouped by area with severity. Triage, not decisions. Triggers: /cla:feedback, 'take these notes one at a time', 'capture my feedback'.

Lightweight end-to-end path for a small change without OpenSpec artifacts: plan, implement with docs and tests, one test pass, open the PR, one review round with a single fix round (a second, on the fix diff, for a high-risk fix). Stops only on an unresolved test failure. Triggers: /cla:lite-pr, 'quick PR for X', 'small change, just ship it'.

Chains multiple /cla:lite-pr runs extracted from a decision-shaped markdown doc: pulls out every lite-pr-sized candidate, sequences them dependency-first (falling back to document order), confirms the plan once up front, then runs /cla:lite-pr on each in order — under the confirmed merge policy either merging every clean PR as the chain goes (recommended) or merging only a PR its dependents need, and quarantining a failed candidate plus its downstream subtree while continuing with the rest. Designed for long, unattended small-change runs off a single decisions/feedback doc. Triggers on /cla:multi-lite or natural language like 'extract all lite-pr candidates from this decision file and run them in sequence', 'chain these small changes end to end', 'lite-pr everything in this doc'.

Chains multiple OpenSpec changes end-to-end: determines implementation order, runs /cla:spec-to-pr on each change in sequence, enforces that every Critical/Important review finding is actually fixed (never left as a deferred TODO) before moving on, merges each PR a later change needs merged — only on the exact head its review and full test gate covered — and finishes with a full repo cleanup pass. Designed for long, unattended multi-change runs. Triggers on /cla:multi-pr or natural language like 'drive all my openspec changes to PRs', 'chain these changes end to end', 'work through every open change and merge as you go', 'run spec-to-pr on everything'.

Turn a shape-decision decisions file into a batch of OpenSpec change proposals, committed one change at a time, reviewed once as a batch, then opened as one PR. Proposals only, no implementation. Triggers: /cla:multi-spec, 'turn these decisions into changes', 'batch-propose the shaped decisions'.

Start an isolated git worktree for this repo with dependencies installed and gitignored env files carried over; inside an existing worktree it runs only the setup half. Triggers: /cla:new-worktree, 'new worktree', 'work on this in a worktree'.

CTO-level review of this repo — assesses vision, structure, requirements, architecture, and validation across the whole codebase. Triggers on /cla:project-review or natural language like 'review the whole project', 'CTO-level review', 'assess the codebase'.

File a defect in the cla plugin's own portable core (a synced skill, agent, hook or script that would be wrong in every repo) as an issue against the canonical source. Triggers: /cla:report-upstream, 'this hook is broken upstream', 'report this to the cla source'.

Pre-implementation review of an OpenSpec change: verify its claims and file and symbol references, size it, then review directly (small) or with three parallel agents (large). Triggers: /cla:review-change, 'review the openspec change', 'check the change before I implement it'.

Recommend the cheapest model and effort that can plausibly do a described task well, then offer to start it with those settings. Run with /cla:right-model.

Persist every tool permission granted this conversation to .claude/settings.local.json. Run with /cla:save-permissions.

Shape a decision by walking through choices one at a time with pros/cons and a recommended pick. Triggers on /cla:shape-decision or natural language like 'help me decide between X and Y', 'shape this decision', 'walk me through the options'.

Drive one OpenSpec change end-to-end, from idea, change name or /opsx:explore result, to an opened, archived PR with review fixes applied. No merge, no deploy. Triggers: /cla:spec-to-pr, 'take this change to a PR', 'spec-to-pr the X feature'.

Review recent /cla:spec-to-pr runs across the fleet's spec-to-pr ledgers (warn reasons, round-cap exhaustion, per-agent Revise yield, ask answers, flag use, diagnose escalations, Revise round-2 yield) and propose edits to the orchestrator. Run with /cla:spec-to-pr-retro.