agent-design-language

Construct all six C-SDLC v2 cards and canonical issue state from typed input.

Record terminal truth and safely close out a green integrated issue.

Assign and record exact-revision pre-publication review truth.

Lightweight sprint orchestrator for ADL. Use when one sprint issue should drive an explicit child issue wave through existing lifecycle and editor skills with a Sprint Execution Packet that records sequential, parallel, or hybrid intent, robust sprint-end review, and truthful sprint closeout.

Lightweight conductor for the ADL workflow skills. Use when the operator wants one bounded entrypoint that detects the current issue/workflow state, selects the correct next lifecycle or editor skill, enforces skill/subagent policy, and either stops after routing or dispatches one bounded downstream skill subtask without reimplementing the underlying work.

Bind a prepared C-SDLC v2 issue to a claim, branch, and worktree.

Apply typed semantic card operations through the canonical v2 record.

Diagnose canonical v2 state without mutation or network access.

Publish only after current pre-publication review truth.

Classify scheduler, validation, review, publication, and readiness next actions.

Execute declared PVF validation DAGs with typed evidence.

Execute a prepared issue after doctor review. Use when an issue is structurally ready, should bind or confirm its execution branch and worktree, then perform the bounded implementation work, validations, and truthful output recording without silently finishing or janitoring the PR.

Watch one issue, PR, branch, or dependency gate for readiness while work is waiting, classify the state, route PR blockers to pr-janitor, and stop before merge, closeout, or implementation.

Finalize an issue after its implementation path is done by verifying GitHub issue/PR closure state, normalizing final STP/SIP/SOR truth, reconciling root/worktree artifacts, recording supersession or duplicate links when relevant, and pruning the local execution surface. Use when publication/review is over or a no-PR closure disposition is known and the next step is truthful local closeout rather than implementation or PR monitoring.

Perform truthful closeout for an executed issue by validating the output record, staging the intended paths, creating or updating the draft PR, and stopping before silent merge. Use when bounded issue execution is complete and the next step is reviewable PR publication or update.

Watch a PR after draft creation, diagnose failed checks or merge conflicts, recommend or apply bounded fixes when clearly justified, and stop before unreviewed scope expansion. Use when a PR is in flight and the user wants help monitoring CI, conflicts, review state, or merge-readiness.

Findings-first review orchestrator for one completed sprint, mini-sprint, issue wave, or release-tail bundle. Use when umbrella issue truth, child issue closure truth, PR state, lifecycle cards, validation evidence, closeout artifacts, and follow-up routing need one consistent review packet composed from the existing ADL review skills rather than a hand-assembled process.

Normalize and correct an SIP input card so it reflects truthful lifecycle state without creating execution context or inventing progress. Use when a SIP has branch/worktree drift, stale target surfaces, placeholder leakage, or readiness-blocking card errors.

Normalize and correct an SOR output card so it records truthful execution and integration state without inventing validation or overstating completion. Use when an output card has placeholders, wrong integration wording, stale validation claims, or finish-blocking truth-model errors.

Normalize and correct an SPP planning card so it preserves truthful pre-execution planning state, valid `codex_plan` statuses, and issue-local planning boundaries. Use when an `spp.md` has stale plan structure, invalid `codex_plan` values, missing stop conditions, weak source references, or readiness-blocking planning drift.

Normalize and correct an SRP review card so it preserves Structured Review Prompt semantics, review-result truth, finding dispositions, and residual-risk boundaries. Use when an `srp.md` is missing, still uses legacy Structured Review Policy scaffolding, lacks review results, or overclaims review completion.

Normalize and tighten an STP task card without changing issue intent or taking over lifecycle orchestration. Use when an STP needs bounded structural cleanup, clearer acceptance criteria, or repo-consistent prompt sections before readiness review or execution.

Plan, write, audit, repair, or polish bounded repository documentation from explicit source evidence. Use when Codex needs to work on README, milestone, feature, ADR, demo, review, architecture, onboarding, runbook, or skill documentation while separating source-backed facts from recommendations and assumptions, flagging stale commands, overclaims, missing validation evidence, and stopping before publication or broad unbounded rewrites.

Diagnose PR workflow readiness and drift for an issue, task bundle, branch, or worktree. Use when the user wants a doctor-style readiness result, wants to know whether workflow state is broken or incomplete, or wants small bounded mechanical repairs applied automatically without widening into implementation work.

Create, split, or reallocate ADL milestones with the full C-SDLC planning package, conductor-bound setup issue, version labels, issue migration truth, feature docs, validation, review, and PR publication. Use when the operator asks to create a new ADL milestone, bridge milestone, next-milestone package, milestone split, milestone setup wave, or to move open issues between ADL versions.

Analyze and manage PR stack topology, dependency order, and base alignment for ADL issue workflows in a bounded, reviewable way.

Initialize a tracked issue for the PR lifecycle. Use when the user wants to create or resolve a GitHub issue, generate the canonical local source issue prompt plus initial STP/SIP/SOR bundle in the correct locations, validate the init surfaces, and stop before any branch or worktree creation.

Review an entire repository or large repo slice for bugs, regressions, security risks, correctness issues, maintainability risks, misleading diagnostics, and missing tests across all severity levels. Use when the user asks for a code review of a whole repo, wants repo-wide risk assessment before merge or release, or needs findings-first review output across multiple files.

Compare an explicit expected baseline against observed implementation, docs, tests, review, PR, milestone, or closeout evidence and produce findings-first gap reports with severity, evidence, uncertainty, and bounded follow-up recommendations without fixing, approving, or mutating repositories.

Normalize and correct ADL milestone or project planning documents that are generated from, intended for, or being aligned with versioned planning templates without editing C-SDLC issue cards or inventing review, approval, release, PR, or closeout truth. Use when README, WBS, sprint, vision, design, decision, demo matrix, checklist, release plan, release notes, or handoff docs have placeholder residue, stale planning claims, missing required sections, or drift from the available planning-template contract.

Evaluate CodeBuddy review packets and reports against the third-party-review quality bar for evidence quality, severity accuracy, actionability, duplication, unsupported claims, specialist coverage, template compliance, residual risk clarity, and publication safety before customer-facing use.

Classify whether one issue should stay intact, split into bounded follow-ons, defer splitting, or stop for operator judgment while preserving issue-graph and card truth.

Classify whether an issue should stay actionable, close as duplicate or superseded work, fold into another issue, or no-op close out with evidence-backed routing.

Specialist docs reviewer for a multi-agent repository review. Use when a review packet needs a bounded documentation role focused on misleading docs, stale commands, onboarding gaps, release and demo truth drift, API/CLI contract drift, and whether documentation overclaims behavior without editing docs.

Synthesis reviewer for a multi-agent repository review. Use when specialist code, security, test, and docs review artifacts need to be merged into one findings-first repo review without hiding severity, disagreement, missing coverage, residual risk, or role-specific caveats, and without claiming merge approval or remediation.

Run one named demo in a bounded way and classify the result as proving, non-proving, skipped, or failed without absorbing the demo's own logic. Use when the user wants consistent demo execution and proof classification rather than manual after-the-fact interpretation.

Assemble milestone release proof surfaces into a bounded reviewable evidence package covering issue and PR evidence, demos, reviews, remediation, validation, non-claims, and residual risks without approving releases, publishing notes, tagging, merging, or closing issues.

Scan bounded repo contracts, tests, packets, and validation surfaces for machine-local assumptions such as absolute host paths, brittle worktree names, temp paths, stale hard-coded contract references, and environment-specific assertions, then report findings or apply only explicit safe mechanical normalizations.

Inspect review packets, review plans, finding registers, demo or proof registers, and milestone review docs for structural readiness, classifying safe cleanup, blockers, skipped surfaces, and follow-on needs without remediating findings, publishing reports, or approving review readiness.

Build a bounded, source-grounded repository review packet for CodeBuddy-style specialist review lanes before any review, diagram, test, issue, or report skill runs.