Total Skills
21
Skills published by KlausFreiberufler with real stars/downloads and source-aware metadata.
Total Skills
21
Total Stars
21
Total Downloads
0
Comparison chart based on real stars and downloads signals from source data.
devflow-code-critic
1
devflow-plan-critic
1
devflow-adr-compliance
1
devflow-area-ideation
1
devflow-collision-acknowledged
1
devflow-core
1
devflow-debugging
1
devflow-error-investigator
1
Use during in_progress state, before submitting flow_update({currentState: 'review'}), to deeply review the implementation against 7 quality dimensions. The skill enforces a critic-persona — the agent steps out of "implementer" mode and into "reviewer" mode. Outputs a structured verdict (approved | minor_issues | needs_changes) with severity-tagged findings. Iron Law: a verdict is only valid if all 7 dimensions were explicitly checked, and the diff plan-vs-reality was actually examined.
Use during planning state, before submitting flow_update({currentState: 'approval'}), to deeply review the plan against 7 quality dimensions. The skill enforces a critic-persona — the agent steps out of "author" mode and into "reviewer" mode. Outputs a structured verdict (approved | minor_issues | needs_changes) with severity-tagged findings. Iron Law: a verdict is only valid if all 7 dimensions were explicitly checked.
Use during the review state of a flow with file changes. Compares the diff against accepted ADRs' affects_paths globs, persists per-file violations, and emits a discipline-token only when every violation is either resolved or explicitly marked break-by-design with a reason ≥ 10 characters. Implements Knowledge-Hook 6 of the Knowledge-Gated Workflow.
Use when the user pasted an Area-Idea-Hunt-Prompt from the IdeasPage Garage. Drives a Wiki-grounded ideation routine — never out-of-thin-air-brainstorming, every idea must reference real wiki entries. Output is 3 Pick&Plan-ready proposals as JSON.
Use during planning state to walk the user through the awareness list (pending_work + planning_context) so every in-flight flow, open intent, proposed ADR, and pending draft in the topic-scope receives an explicit user decision (✓ keep / ✗ irrelevant / ↺ later). Emits a discipline-token at the end as proof for agent_with_discipline self-approval.
Use whenever working in a DevFlow-managed project (CLAUDE.md contains devflow_init). Explains the flow-state machine, init requirements, and error recovery.
Use when an unexpected behavior, failing test, or production-error needs to be diagnosed. Walks through four phases — Investigation → Pattern → Hypothesis → Fix — with an explicit Knowledge-Hook 4 runbook-match before the Hypothesis phase to avoid double-debugging known incidents.
Use FIRST when you encounter an unexpected error / exception / failing test that's not immediately obvious. Calls `error_context_get` to pull all wiki signals relevant to the error (runbooks, ADRs touching the file, recent flows, patterns, similar past errors). Iron Law of compounding wikis: every novel failure mode produces a new runbook on the way out.
Use when the active DevFlow flow is in 'in_progress' state. Guides implementing the approved plan with tasks, commits, and TDD discipline.
Use whenever listing flows, presenting flow status, or summarizing the flow backlog. Enforces a uniform Markdown-table format across MCP-tool, slash-commands, and free-form agent responses so the user gets a consistent view everywhere.
Use periodically (monthly cadence is plenty) to curate the project's idea backlog. Reads `GET /api/projects/:id/ideas` — five sources of organic ideas (intents, stale, orphans, contradictions, hotspots). For each item: pick & plan into a flow, defer with a substantive horizon, or reject with a reason. Iron Law: no idea sits unaddressed for more than 90 days.
Use during planning, before submitting to approval, and again before review→done. Walks the gaps from `wiki-context` and ensures each gap is closed with `extend` (preferred), a new entry, or a deferred intent — never `dismiss`. The wiki is a compounding artifact: knowledge is never thrown away.
Use during the planning state to surface existing project patterns that match the flow's tag-scope, so the implementation plan re-uses proven solutions instead of inventing parallel ones. Cites each accepted pattern via `[[pattern:slug]]` wiki-link in the plan.
Use during the review state, before devflow-verification-gate, to reconcile the approved implementation_plan against what was actually implemented. Produces a per-AC and per-section table — done / partial / missing / extra — and forces the agent to address every gap (either fix it now, file an intent, or document why it's deferred). Without this skill the review only proves AC-by-AC, not that the plan as a whole was respected.
Use when the active DevFlow flow is in 'planning' state. Guides creating an implementation plan with acceptance criteria before submitting for approval.
Use when receiving a code-review (from a subagent dispatch or a human reviewer) to triage every finding into Critical / Important / Minor with a deliberate decision per item. Each finding is technically verified before accept-or-reject — never blind acceptance, never blind rejection.
Use when the active DevFlow flow is in 'review' state. Guides self-review via diff analysis and writes the agent summary + testing instructions.
Use during implementation to drive every change through Test-Driven Development. Before writing production code, query existing test-patterns from the LLM-Wiki (Hook 3) for the relevant tag-scope, then write a failing test (RED), make it pass (GREEN), and refactor (REFACTOR). Emits a discipline-token only when each cycle has the prescribed evidence.
Use during the review state to prove every acceptance criterion with a captured evidence command. No "should work" claims — each AC gets a real shell/test/curl invocation whose stdout is recorded into the flow's testingInstructions. Emits the discipline-token that makes review→done possible under agent_with_discipline self-approval (DF-292).
Use periodically (or on demand from the Wiki page health-bar) to keep the LLM-Wiki clean. Reads `GET /api/projects/:id/wiki-lint` and addresses every reported stale, orphan, and contradiction — never let any of them linger. Iron Law: orphans get cross-linked or marked as `intent`, contradictions are resolved at the ADR-status level, stale entries are reviewed and either extended or superseded.
Use at the start of implementation (or when picking up an existing flow) to set up a dedicated git-worktree under .claude/worktrees/<displayId>-<slug>. Each flow gets its own working directory — branch isolation prevents cross-flow contamination, and the path is predictable for re-init / cross-session recovery.