Destynova2/cli-code-skills / cli-cycle
cli-cycle
UnverifiedContinuous improvement cycle — orchestrates all cli-* skills on the current project, synthesizes results, and proposes prioritized improvements. Use when the user wants a full project review, a health check, a weekly cycle, or says 'audit everything', 'review the project', 'health check', 'what should I improve', 'run all audits', 'cycle', 'improvement cycle'. Designed for recurring use with '/loop 7d /cli-cycle'.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 13.8 KB |
| references/convergence.md | markdown | 10.8 KB |
| references/orchestration.md | markdown | 22.1 KB |
| references/skill-flow.md | markdown | 11.1 KB |
references/skill-flow.md
View on GitHub# Skill Flow Graph — Directed Acyclic Graph of Skill Triggers
> **When to read:** During Steps 4-6 when deciding execution order, handling handoffs, and triggering forge skills from audit findings.
---
## Contents
- Flow philosophy (inspired by cli-audit-tangle)
- The DAG
- Trigger edges (who triggers whom)
- Deduplication matrix
- Cycle detection (safety)
- Execution order summary
- Forge skill modes
---
## Flow philosophy (inspired by cli-audit-tangle)
The skill ecosystem is a **directed graph**. Each edge represents a trigger condition:
- `audit-X --[condition]--> forge-Y` = audit finding triggers a forge correction
- `audit-X --[handoff]--> audit-Y` = audit recommends deeper analysis by another audit
- `forge-X --[always]--> git-conventional` = every forge output gets committed
**Anti-pattern detection:** If the graph has cycles, skills call each other forever. We use the same Tarjan SCC detection as tangle: any cycle = bug in the flow.
---
## The DAG
```
┌─────────────┐
│ cli-cycle │
│ (orchestrator)│
└──────┬───────┘
│
┌─────────────┼─────────────┐
│ │ │
┌─────▼─────┐ ┌────▼────┐ ┌──────▼──────┐
│ WAVE 1 │ │ WAVE 2 │ │ WAVE 3 │
│ (parallel) │ │(parallel│ │ (adaptive │
│ │ │+handoffs│ │ handoffs) │
└─────┬──────┘ └────┬────┘ └──────┬──────┘
│ │ │
└─────────────┼─────────────┘
│
┌──────▼───────┐
│ TRIAGE │
│ 3-2-1 │
└──────┬───────┘
│
┌──────▼───────┐
│ CORRECTION │
│ pipeline │
└──────┬───────┘
│
┌─────────────┼─────────────┐
│ │ │
┌─────▼─────┐ ┌────▼────┐ ┌──────▼──────┐
│ Direct fix │ │ Forge │ │ git-conv │
│ (edit code)│ │ (generate│ │ (commit) │
│ │ │ docs) │ │ ALWAYS │
└────────────┘ └─────────┘ └─────────────┘
│
┌──────▼───────┐
│ RE-AUDIT │
│ (affected │
│ skills only)│
└──────┬───────┘
│
converge?
┌──▼──┐
│ oui │──→ FIN
│ non │──→ CORRECTION (boucle)
└─────┘
```
---
## Trigger edges (who triggers whom)
### Wave 1 → Wave 2 (Handoff edges)
Each edge fires only if the condition is detected during Wave 1.
```
cli-audit-code ──[god modules detected]──────────→ cli-audit-tangle (Wave 2)
cli-audit-code ──[hidden semantic cost]──────────→ cli-audit-xray (Wave 2)
cli-audit-code ──[C9 security issues in .sh]─────→ cli-audit-shell (already Wave 1, inject reason)
cli-audit-code ──[C8 test quality low]───────────→ cli-audit-test (already Wave 1, inject reason)
cli-audit-shell ──[sed on YAML detected]─────────→ cli-forge-infra (Wave 2)
cli-audit-shell ──[N scripts, no getopts]────────→ cli-audit-wizard (already Wave 1, inject reason)
cli-audit-shell ──[scripts are CI entrypoints]───→ cli-forge-pipeline (Wave 2)
cli-audit-tangle ──[CI deadlocks detected]───────→ cli-forge-pipeline (Wave 2)
cli-audit-tangle ──[god/boundary funcs mix dataflow + resources]→ cli-audit-xray (Wave 2)
cli-audit-tangle ──[call graph > 20 nodes]───────→ cli-forge-schema (Wave 2)
cli-audit-doc ──[docs describe dead features]────→ cli-audit-sync (Wave 2)
cli-audit-doc ──[README score low]───────────────→ cli-forge-readme (already Wave 1, inject reason)
cli-audit-test ──[no drift detection D13]────────→ cli-audit-drift (Wave 2)
cli-audit-test ──[CI has no test stage]──────────→ cli-forge-pipeline (Wave 2)
cli-audit-code ──[SQLx transaction/data red flags]→ cli-audit-data (Wave 2)
cli-audit-test ──[PostgreSQL proof gaps]──────────→ cli-audit-data (Wave 2)
cli-audit-hanoi ──[displaced responsibilities across layers]→ cli-forge-infra (Wave 2)
cli-audit-shell ──[wait-for/sleep/sed hacks in entrypoints]→ cli-audit-hanoi (already Wave 1, inject reason)
```
### Wave 2 → Wave 3 (Adaptive handoff edges)
```
cli-audit-sync ──[shell commands in docs broken]──→ cli-audit-shell (if not run in Wave 1)
cli-audit-sync ──[stale diagrams]──────────────────→ cli-forge-schema (inject reason)
cli-audit-drift ──[god-level contracted funcs]─────→ cli-audit-tangle (inject reason)
cli-audit-xray ──[benchmarkable candidate]─────────→ cli-forge-perf (if not run)
cli-audit-xray ──[semantic graph worth visualizing]→ cli-forge-schema (inject reason)
cli-forge-pipeline ──[shell entrypoints in CI]─────→ cli-audit-shell (if not run)
cli-forge-infra ──[complex deployment, no HLD]─────→ REPORT ONLY (forge-hld is Skip)
cli-audit-data ──[unsafe DB design needs correction]→ REPORT ONLY (forge-data is correction-only)
```
`cli-audit-xray` may recommend `cli-audit-drift`, `cli-audit-test`, or `cli-audit-tangle` in its human report when an optimization needs invariants, equivalence tests, or structural isolation. Under `cli-cycle`, those are report-only recommendations if the target audit already ran in an earlier wave; do not create a backward automatic edge.
### Triage → Correction (Forge trigger edges)
These edges fire during the Phoenix correction phase, based on triage items.
```
Finding: "missing CONTRIBUTING.md" ────→ cli-forge-doc
Finding: "missing/outdated README" ────→ cli-forge-readme
Finding: "missing diagrams" ───────────→ cli-forge-schema
Finding: "CI missing stages" ──────────→ cli-forge-pipeline (generation mode)
Finding: "unsafe PostgreSQL/SQLx design" ─→ cli-forge-data
Finding: "code/config issue" ──────────→ Direct edit (no forge skill needed)
ALL corrections ───────────────────────→ cli-git-conventional (ALWAYS)
```
---
## Deduplication matrix
When multiple edges point to the same skill, apply tangle-aware deduplication:
| Same skill | Same scope | Action |
|-----------|-----------|--------|
| cli-audit-tangle | null + null | **Merge** — run once, combine reasons |
| cli-audit-shell | `setup.sh` + `deploy/*.sh` | **Scoped** — run once on `deploy/` (broader scope wins) |
| cli-forge-pipeline | null + null | **Merge** |
| cli-forge-schema | null + null | **Merge** |
| cli-audit-code | `src/core/` + `src/api/` | **Scoped** — run twice, different targets |
| Any skill | Already ran in earlier wave | **Skip** — inject reason only |
**Scope resolution:** When merging scoped calls, take the **broader scope** (directory > file > function). If scopes don't overlap, run separately.
---
## Cycle detection (safety)
The flow graph MUST be acyclic. Apply Tarjan SCC check on the edge list:
**Known safe patterns:**
- `audit-X → forge-Y → git-conventional → STOP` (linear, no cycle)
- `audit-X → audit-Y → forge-Z` (forward only, no cycle)
- `re-audit → same audit skills` (this is the Phoenix LOOP, not a graph cycle — it's controlled by convergence criteria)
**Forbidden cycles:**
- `audit-X → audit-Y → audit-X` (infinite handoff ping-pong)
- `forge-X → audit-Y → forge-X` (fix creates a finding that triggers the same fix)
**Prevention rules:**
1. A handoff edge can only point **forward** (Wave 1 → Wave 2 → Wave 3, never backward)
2. A skill can recommend itself ONLY with a **different scope** (not the same file)
3. Wave 3 is terminal — no Wave 4. Remaining handoffs go to the Recommendations section
4. During correction, if a fix triggers a new finding that would re-trigger the same forge skill → STOP, flag as divergence
---
## Execution order summary
```
Phase 1: AUDIT
Wave 1 (parallel): code, doc, test, tangle, tree, readme, shell, wizard
↓ collect handoffs
Wave 2 (parallel): sync, drift, pipeline, schema, infra + handoff-added skills
↓ collect handoffs
Wave 3 (if needed): handoff-only skills not yet run
↓
Triage 3-2-1
Phase 2: CORRECTION (for each triage item)
Fix item → forge if needed → git-conventional commit
↓
Re-audit (affected skills only)
↓
Converge? → YES: done | NO: next item or next pass
Phase 3: REPORT
Score card + triage + strengths + trends + convergence status
```
---
## Forge skill modes
Each forge skill operates in 2 modes depending on context:
| Mode | Triggered by | Output |
|------|-------------|--------|
| **Audit mode** (Wave 1-3) | Applicability matrix condition | Score + findings (read-only) |
| **Correction mode** (Phoenix fix) | Triage item requiring generation | Generated files (write) |
Skills that operate in both modes:
- `cli-forge-readme` — audit: scores README (RCI). correction: rewrites README
- `cli-forge-tree` — audit: scores structure (SHS). correction: reorganizes files
- `cli-forge-pipeline` — audit: scores pipeline (PSI). correction: adds/fixes CI jobs
- `cli-forge-schema` — audit: checks existing diagrams. correction: generates new diagrams
- `cli-forge-infra` — audit: scores infra (IDS/DHS). correction: simplifies config
Skills that operate in correction mode only:
- `cli-forge-doc` — generates CONTRIBUTING.md, architecture docs, troubleshooting
- `cli-forge-data` — designs and implements corrections for database-safety findings
- `cli-git-conventional` — formats all commits (always active during correction)
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| rea-analysis-optimizationmorluto/reaOptimize REA providers and shared analysis workflows for agent context, latency, CPU, and memory while preserving correctness and Evidence. Use for analysis bottlenecks and cache or query design; CI/test-suite optimization belongs to rea-ci-tests-audit. | Stars: 467k | Updated | Claude CodeCodex | Verified | Install |
| rea-ci-tests-auditmorluto/reaAudit REA CI and test suites for correctness, fixture drift, false passes, and measured performance; use for suite or workflow investigations and efficiency changes, rather than ordinary feature testing or a single test failure. | Stars: 467k | Updated | Claude CodeCodex | Verified | Install |
| rea-tool-designmorluto/reaAudit, investigate, design, or change REA investigation tools, CLI/MCP contracts, provider capabilities, and Evidence semantics. | Stars: 467k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-auditDietrichGebert/ponytailQuality audit of the whole repo: bugs, security, real load, missing tests, speed, and what to delete. Most important first. | Stars: 193k | Updated | Claude CodeCodex | Verified | Install |