Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Initialize plugin in a project — install Iron Laws, auto-activation rules, and reference auto-loading into CLAUDE.md. Use when setting up or updating the plugin.
.claude/skills/oliver-kriska-init/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-14 | ✗→✓ | ▲ Improved | 20% | 0% |
| case-08 | ✓→✗ | ▼ Worse | 152% | 0% |
| case-10 | ✓→✗ | ▼ Worse | -11% | 0% |
| case-11 | ✓→✗ | ▼ Worse | -18% | 0% |
| case-13 | ✓→✗ | ▼ Worse | 39% | 0% |
Install the Elixir/Phoenix plugin's behavioral instructions into the project's CLAUDE.md.
/phx:init # First-time installation
/phx:init --update # Update existing installation with latest rulesUse Glob to check if CLAUDE.md exists. Then use Grep to check for existing ELIXIR-PHOENIX-PLUGIN:START marker in CLAUDE.md.
Scan the project to customize the injected instructions:
Read mix.exs and use Grep to extract:
phoenix.*"~> in mix.exsecto.*"~> in mix.exs"oban" and "oban_pro" in mix.exs"ash", "surface" in mix.exs"tidewave" in mix.exslib/**/*.ex filesMode A: Fresh Install (no CLAUDE.md or no markers)
Mode B: Update (--update flag or markers exist)
<!-- ELIXIR-PHOENIX-PLUGIN:START --> and <!-- ELIXIR-PHOENIX-PLUGIN:END -->CRITICAL: NEVER overwrite or delete existing CLAUDE.md content outside the plugin markers — user-written rules, project conventions, and other plugin sections must be preserved verbatim
Write the following structure to CLAUDE.md:
markdown<!-- ELIXIR-PHOENIX-PLUGIN:START --> <!-- Last updated: {date} | Plugin version: 1.0 | Stack: Phoenix {version}, Ecto {version}, {optional: Oban, Tidewave} --> # Elixir/Phoenix Plugin - Auto-Activation Rules {Include all sections from the Content Template below, filtered by detected stack} <!-- ELIXIR-PHOENIX-PLUGIN:END -->
If command -v codex succeeds (or the user asks): offer to install a managed ## Review guidelines block into the project's AGENTS.md — honored by BOTH codex exec review locally and the Codex cloud PR reviewer. Follow the install rules and exact block in ${CLAUDE_SKILL_DIR}/references/codex-review-guidelines.md: markers ELIXIR-PHOENIX-REVIEW-GUIDELINES:START/END, upsert in place on --update, never touch content outside markers, and STOP to ask if an unmanaged ## Review guidelines section already exists. When codex is absent and not requested, skip silently — never mention codex to users who don't have it.
✅ Elixir/Phoenix plugin initialized
Detected stack:
- Phoenix {version}
- Ecto {version}
- {Oban standard | Oban Pro | not detected}
- {Tidewave ✓ | Tidewave not detected}
- {Ash Framework detected - Ecto patterns disabled | not detected}
Added to CLAUDE.md:
- Auto-activation rules (complexity detection, interview mode)
- Agent trigger patterns ({n} agents available)
- Reference auto-loading ({n} reference docs)
- Iron Laws enforcement ({n} laws)
- Verification rules
{- Codex review guidelines → AGENTS.md (only if installed in Step 4b)}
Run /phx:init --update after plugin updates.
Run /phx:audit for a full project health check.The exact content to inject is in ${CLAUDE_SKILL_DIR}/references/injectable-template.md.
Key structure:
/phx:* command fits which kind of request{OBAN_SECTION} — If Oban detected (not Pro){OBAN_PRO_SECTION} — If Oban Pro detected{ASH_SECTION} — If Ash Framework detected{TIDEWAVE_SECTION} — If Tidewave detectedPlaceholder substitution:
| Placeholder | Source | |-------------|--------| | {DATE} | Current date | | {PHOENIX_VERSION} | From mix.exs | | {ECTO_VERSION} | From mix.exs | | {OPTIONAL_STACK} | Detected optional deps |
See ${CLAUDE_SKILL_DIR}/references/injectable-template.md for full template with all placeholders and conditional sections.
After running /phx:init:
| Scenario | Action | |----------|--------| | CLAUDE.md read-only | Error: "Cannot modify CLAUDE.md - check permissions" | | Markers corrupted | Warn, offer to remove and reinstall | | Unknown Phoenix version | Use conservative defaults (all features enabled) | | Not an Elixir project | Error: "No mix.exs found - is this an Elixir project?" |
| Command | When to Use | |---------|-------------| | /phx:init | First time, or after plugin updates | | /phx:audit | Periodic project health check | | /phx:verify | After code changes |
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 4,685 | 5,481 | +17% | 1 | 1 | 0% | 223 | 1,839 | +725% | 0 | 0 | — |
case-02 | fail→fail | 15,993 | 6,182 | -61% | 1 | 1 | 0% | 2,581 | 1,880 | -27% | 0 | 0 | — |
case-03 | fail→fail | 4,475 | 6,689 | +49% | 1 | 1 | 0% | 205 | 1,872 | +813% | 0 | 0 | — |
case-04 | fail→fail | 12,102 | 5,687 | -53% | 1 | 1 | 0% | 2,030 | 1,750 | -14% | 0 | 0 | — |
case-05 | fail→fail | 3,068 | 6,459 | +111% | 1 | 1 | 0% | 230 | 1,870 | +713% | 0 | 0 | — |
case-06 | fail→fail | 14,011 | 20,520 | +46% | 1 | 1 | 0% | 2,398 | 1,773 | -26% | 0 | 0 | — |
case-07 | fail→fail | 8,963 | 5,563 | -38% | 1 | 1 | 0% | 1,378 | 1,833 | +33% | 0 | 0 | — |
case-08 | pass→fail | 4,496 | 6,972 | +55% | 1 | 1 | 0% | 749 | 1,890 | +152% | 0 | 0 | — |
case-09 | fail→fail | 2,944 | 5,330 | +81% | 1 | 1 | 0% | 440 | 1,810 | +311% | 0 | 0 | — |
case-10 | pass→fail | 13,273 | 5,378 | -59% | 1 | 1 | 0% | 1,948 | 1,734 | -11% | 0 | 0 | — |
case-11 | pass→fail | 12,753 | 6,444 | -49% | 1 | 1 | 0% | 2,173 | 1,785 | -18% | 0 | 0 | — |
case-12 | fail→fail | 16,447 | 6,478 | -61% | 1 | 1 | 0% | 2,645 | 1,900 | -28% | 0 | 0 | — |
case-13 | pass→fail | 7,795 | 4,951 | -36% | 1 | 1 | 0% | 1,330 | 1,847 | +39% | 0 | 0 | — |
case-14 | fail→pass | 10,716 | 4,179 | -61% | 1 | 1 | 0% | 1,806 | 2,162 | +20% | 0 | 0 | — |
case-15 | fail→fail | 2,642 | 5,473 | +107% | 1 | 1 | 0% | 252 | 1,844 | +632% | 0 | 0 | — |
case-16 | fail→fail | 12,314 | 5,884 | -52% | 1 | 1 | 0% | 1,991 | 1,849 | -7% | 0 | 0 | — |
case-17 | pass→fail | 11,856 | 4,931 | -58% | 1 | 1 | 0% | 2,022 | 1,781 | -12% | 0 | 0 | — |
case-18 | pass→fail | 7,064 | 6,504 | -8% | 1 | 1 | 0% | 1,319 | 1,800 | +36% | 0 | 0 | — |
case-19 | pass→fail | 8,621 | 4,833 | -44% | 1 | 1 | 0% | 1,356 | 1,682 | +24% | 0 | 0 | — |
case-20 | pass→fail | 16,078 | 25,072 | +56% | 1 | 1 | 0% | 1,725 | 1,769 | +3% | 0 | 0 | — |
case-21 | pass→pass | 5,354 | 3,706 | -31% | 1 | 1 | 0% | 897 | 1,880 | +110% | 0 | 0 | — |
case-22 | fail→fail | 9,501 | 4,957 | -48% | 1 | 1 | 0% | 1,830 | 1,778 | -3% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 2 counted toward the lift figure. The other 20 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of -32 percentage points is the difference between those two pass rates over the 2 comparable cases. 12 cases got worse with the skill loaded, and they are included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.