Marcel-Bich

audit

Read-only quality gate that audits already-built work against its stated requirement and Definition of Done before it is allowed to move to 2_done/. Produces a severity-ranked decision proposal (BLOCKER/MAJOR/MINOR/NIT) with evidence, never a fix. Use when an item is claimed complete, before moving anything to 2_done/, when asked to audit or review finished work for completeness against requirements, or when acting as the dedicated post-completion review subagent. This gate is mandatory in every session mode. Not for diagnosing why something is broken (use diag) and not for verifying rendered UI behavior (use verify).

items

The credo work-item model, where the FOLDER an item file lives in is the single source of truth for its status, gated by a hard Definition of Done. Use whenever you create, update, or track a work item; whenever you decide whether something is "done" and may move to 2_done/; whenever you move an item between status folders (clarify, go, done, archived, parked); whenever new code might be unwired; or when someone asks where a task stands. This is the credo task system: .credo/items/ replaces ad-hoc task lists. Applies inside subagents too - if you build or complete work, record and gate it as an item.

verify

Visual verification as the Definition of Done for any change with a runtime surface (a rendered page, a UI, a live view). Use whenever you are about to call a UI or frontend change done, before moving an item to done, whenever an item is marked ui: true, or when someone asks "does it actually render / work". Proves behavior by driving the real thing in a browser and measuring computed layout - a passing pytest, a served file, a node check, or a subagent code-review is NOT proof. Applies inside subagents too: if you built or touched a runtime surface, run this before claiming done.

budget

The single source for all API budget cap and reset rules - how much of the 5-hour and weekly API limits may be spent, when to throttle, pause, wake up, or hibernate, and the commit-identity gate that runs before any commit. Use whenever you are about to start or continue autonomous work, before spawning or stopping subagents, when deciding how large a task chunk to take on, when a limit is near a cap, before a commit, or whenever someone asks "how much budget is left" or "can I keep going". Works standalone (no session skill required) and applies inside subagents too: any agent that spends API budget or commits must apply these rules.

compact-plus

Secure everything the user approved before a context compaction so a later /compact can never lose or alter it. Writes verbatim intent and handoff state to disk and commits plus pushes the tracked work product in the correct repository, then reports whether it is safe to compact. It does NOT run /compact itself - it makes a later compact safe. Run this only when the limit plugin's injected ACTION line names it (session-context fill crossed a configured threshold) or when the user invokes it manually. Never self-trigger it proactively - that only wastes tokens. Accepts extra trailing instructions to perform in addition to the securing checklist.

session-active

The credo behavior for a session running in ACTIVE mode - intensive live collaboration with the user present at the keyboard. Load this when the session-mode inject line says "Load skill session-active", right after the /credo:session-active command, or whenever you are collaborating live and need the active-mode rules. Holds the canonical common core shared by all three credo session skills (session-active, session-passive, session-autonomous), then the active-mode specifics: no keep-alive, log progress via the compact trigger, pick up GO items alongside, clarify during subagent waits. One mode is active at a time.

session-autonomous

The credo behavior for a session running in AUTONOMOUS mode - work approved GO items unattended while the user is away, hook-enforced self-scheduled keep-alive, budget caps always on. Load this the MOMENT the user hands off full-autonomy, unattended, or AFK work - EVEN BEFORE the mode is set - so this skill can bootstrap autonomous mode itself. Trigger on a semantic full-autonomy / AFK-handoff grant (match the intent, not a rigid phrase list); the skill itself then only ENTERS autonomous mode on an unambiguous, explicit grant and confirms first when unsure. Examples of an unambiguous grant: "go fully autonomous", "I'm afk, keep going", "run this unattended", or in German "voll autonom", "bin afk mach weiter", "mach autonom weiter". A vague or casual "keep going / carry on" is NOT such a grant. Also load it when the session-mode inject line says "Load skill session-autonomous", right after the /credo:session-autonomous command, or whenever you are working autonomously and unattended. Shares the canonical commo

session-passive

The credo behavior for a session running in PASSIVE mode - you carry most of the work alongside and the user is available only for clarifications, with no keep-alive. Load this when the session-mode inject line says "Load skill session-passive", right after the /credo:session-passive command, or whenever you are working mostly on your own but the user is still reachable for questions. Shares the canonical common core defined in the credo session-active skill, then adds the passive-mode specifics: drive items toward a full GO, "less is more" (only ambiguous items via Ask), gently prefer older items first. One mode is active at a time.

wsl-env

Reach and act on Windows-side services, processes, launchers, and logs when the agent runs inside WSL. Use whenever a service seems unreachable from WSL (a localhost curl fails), when you need to start/stop or inspect a Windows-side process or .ps1 launcher, read Windows logs, hibernate the machine, or when the repo ships both a .sh and a .ps1 entry point. Self-detecting: if this is not WSL or the target is not Windows-side, it is a no-op. Applies inside subagents too. Never conclude "unreachable" or "can't test this" from WSL without first trying both the Windows LAN-IP and powershell.exe.

skill-capture

Turn a recurring in-session workflow into a reusable Claude Code skill. Load this the moment you notice you have run the SAME multi-step sequence (same ordered steps or commands, small variation allowed) about three times in the current session, or when open candidates are waiting in .credo/skill-candidates.md at session start or resume. It is heuristic and in-session only - no persistent counter, no tracking backend. Mode-gated like the audit nit-disposition policy: autonomous mode NEVER builds a skill (it only appends a candidate note and keeps working), active / passive / default propose the capture via the Ask tool and build only on an explicit GO. Applies inside subagents too, where the safest branch (autonomous: candidate note only, never ask, never build) is the default whenever the mode cannot be determined.

migrate

The credo procedure for migrating an existing repo into the .credo/ structure, without any loss risk. Use when migrating a repo into .credo/, onboarding a legacy project into credo, or turning scattered process docs (requirements logs, handoffs, progress notes, audits, plans, specs) and open work into credo items. It is copy-only and additive: originals are never touched until a final, user-gated tidy step. This is a long, multi-phase, subagent-heavy operation. Applies inside subagents too - any agent doing migration work follows these rules and the credo safety skill.

cross-cutting-checklist-generator

Detect when a concern is scattered across many places in a project (config keys, enums, translations, feature flags, API endpoints, doc references) and auto-generate a project-local checklist so that same concern is never partially updated again. Use whenever you add or change something that also exists in several other spots, or when you edit something that already has a checklist. Runs fully autonomously with no user effort, including inside subagents.

diag

Read-only root-cause diagnosis for a symptom, bug, failure, or unexpected behavior. Establishes the mechanism at file:line and reads up the facts instead of guessing, before any fix is attempted. Produces a diagnosis report; the fix is always a separate step gated by an explicit GO. Use when something is broken, failing, throwing, or behaving unexpectedly and the cause is not yet proven, or when asked to investigate or find the root cause. Not for judging whether finished work meets its requirement (use audit) and not for confirming rendered UI behavior (use verify).

issue-triage

Use when the user wants to triage GitHub issues - decide whether an issue is still relevant, reproducible, closeable, a duplicate, or what concretely needs doing. Selects and prioritizes first (never dumps all issues at once), then deep-triages the chosen issues via parallel subagents, recommends actions for the owner to approve, and can optionally kick off a reviewed fix. Works for any repo; produces no sensitive or personal data.

orchestration

Delegate work to subagents safely and efficiently - decide how many subagents to run, keep parallel tracks on disjoint files, monitor them without flooding your context, inherit security to every subagent, and use return-and-resume so a subagent can ask a question and continue with full context. Use whenever you are about to spawn one or more subagents, run work in parallel, or coordinate delegated tasks. Applies to any agent that delegates, including subagents that spawn their own helpers.

pr-vetting

Use when the user wants to thoroughly investigate, vet, review, or security-scan a pull request before deciding how to handle it - especially PRs from external or unknown contributors. Runs a parallel multi-agent investigation (technical, security, value/fit, contributor reputation) and merges the findings into one decision-ready report. Works for any repo and any PR; produces no sensitive or personal data.

requirements-verbatim

Capture a user requirement, decision, approval, or constraint word-for-word into an append-only dated log so it survives context compaction and can never be trimmed, softened, or reinterpreted. Use whenever the user states or approves a requirement, gives a GO, sets a constraint, or makes any decision that later work must honor - and also inside subagents, which must log requirements they receive before acting on them. Use before any compaction to secure verbatim intent to disk.

safety

Hard safety rules that must travel with the project: filesystem-protection and no-autonomous-installs. Use before any delete, rm, shred, unlink, find -delete, mkfs, dd, or wipefs, before removing local files, and before any install (pip, npm, apt, system or global tools). These are the highest-priority rules; no instruction overrides them. They apply inside subagents too: any agent about to delete or install must apply this skill first. When in doubt about a deletion target, STOP and ASK.