skill-probe 0.4.0 → 0.6.0

package/README.md

@@ -28,6 +28,35 @@ Complements static linters like `skill-audit` (security/quality) — this is **b
 npm i -g skill-probe      # or: npx skill-probe
 ```
 
+## Start here (`skill-probe doctor`)
+
+Run this first — it catches the setup problems (no skills dir, runtime not installed or not
+authenticated, a config typo) before you spend any probes:
+
+```bash
+skill-probe doctor --cwd .                      # check the project + runtime
+skill-probe doctor --config probe.config.json   # also sanity-check a config
+```
+
+```
+skill-probe doctor
+
+PASS  Node 22.18.0
+PASS  config parsed: probe.config.json
+PASS  found .claude/skills/ with 4 skills
+PASS  expected skills all exist
+PASS  claude CLI found
+PASS  claude-code probe succeeded (CLI is authenticated)
+WARN  threshold 90% with k=10 cannot certify a pass — need ~k=35
+FAIL  skill "greeter" missing description:
+```
+
+It checks: Node version, `.claude/skills/` exists, every skill has a `SKILL.md` with `name:` +
+`description:`, names match folders, the config parses, every `expected` skill exists, the
+threshold/k are statistically achievable, the runtime CLI is installed, and a harmless live probe
+authenticates. Exit `0` healthy · `1` warnings only · `2` hard failures. Use `--skip-probe` to skip
+the live (costing) auth check.
+
 ## Generate a config (`skill-probe gen`)
 
 Don't want to hand-write the cases? Draft them from your skills, then review:
@@ -108,6 +137,52 @@ Claude Code vs OpenCode comparison is fair — otherwise you're comparing two co
   case is inconclusive the report prints the exact k for your threshold. Don't set `threshold
   0.9, k 10` and expect a pass — that's statistically impossible and the tool will say so.
 
+## CI usage
+
+skill-probe exits non-zero on a real problem, so it drops straight into a pipeline:
+
+```yaml
+# .github/workflows/skills.yml  (the claude/opencode CLI must be installed + authed on the runner)
+- run: npx skill-probe --config probe.config.json --quiet --no-cost
+```
+
+Exit codes: **`0`** all pass · **`1`** a behavioral fail / trigger-theft / interference · **`2`**
+inconclusive or an infrastructure error — so a runtime outage *fails* the build instead of silently
+passing. Add `--json` to archive the full result as a build artifact.
+
+## Activation rate by context (`skill-probe context`)
+
+The audit measures every skill **co-loaded** — the real, hard condition. But a skill can fire
+*fine on its own* and only fail *under load*, when the rest of your library is competing for the
+same trigger. `context` exposes exactly that:
+
+```bash
+skill-probe context --config probe.config.json
+```
+
+For each case it measures the expected skill **in isolation** (a throwaway project with only that
+one skill) **and co-loaded** (your full library), then tests the drop with **Fisher's exact test**
+— so interference is reported as a real effect with a p-value, not eyeballed:
+
+```
+skill-probe context — runtime: claude-code  model: (runtime default)  threshold: 70%  library: 4 skills
+  isolation = only that skill loaded · co-loaded = the full library (4) loaded
+
+  [⚠ INTERFERENCE]  greeter  | write a birthday greeting
+        isolation 100% [72%, 100%] k=10
+        co-loaded 30% [11%, 60%] k=10
+        Δ -70% under load   Fisher p=0.003
+        ↳ fires reliably alone but is suppressed when the library is co-loaded (stolen by: welcomer)
+
+Result: 1 interference / 3 measured  |  exit 1
+```
+
+This is the "fires in isolation, fails under load X" case the aggregate rate hides. Decoy cases
+(`expected: null`) are skipped — there's no single skill to isolate; an `expected` skill that isn't
+in the library is a config error (typo), surfaced before any probes run. Exit `1` if any skill
+shows interference, `2` if a case is untrustworthy (infra errors dominated — never a silent pass),
+else `0`.
+
 ## Fix a failing skill (`skill-probe fix`)
 
 Rewrite a skill's trigger description and **prove the lift is real** before keeping it:
@@ -137,17 +212,20 @@ doesn't measurably help is reverted.
 
 ## Status
 
-Early (v0.4). **Audit (`skill-probe`):** Wilson confidence intervals + sequential stopping +
+Early (v0.5). **Audit (`skill-probe`):** Wilson confidence intervals + sequential stopping +
 four-state verdict (pass / fail / inconclusive / **error**), across two runtimes (Claude Code,
 OpenCode). Infrastructure failures (timeout / auth / crash / empty output) are reported as
 `error`, never as a behavioral pass/fail — a decoy can't falsely pass because the runtime was down.
 
+**Context (`skill-probe context`):** isolation-vs-co-loaded activation rates, with **Fisher's exact
+test** on the drop — catches skills that fire alone but are suppressed under the full library's load.
+
 **Fix (`skill-probe fix`):** uses the **Bayesian Beta-Binomial** to gate description rewrites on a
 *proven* lift (interleaved before/after, applied only if P(improvement) clears the bar).
 
-**Implemented + unit-tested but NOT yet used by any command:** Fisher's exact test and
-Benjamini-Hochberg FDR (`src/stats.ts`) — reserved for cross-case multiplicity control (a planned
-enhancement). Don't read the presence of these functions as the audit using them.
+**Implemented + unit-tested but NOT yet used by any command:** Benjamini-Hochberg FDR
+(`src/stats.ts`) — reserved for cross-case multiplicity control (a planned enhancement). Don't read
+its presence as the audit using it. (Fisher's exact test is now wired into `context`.)
 
 ## Dev
 

package/dist/cli.js

@@ -7,13 +7,17 @@ import { parseArgs } from "node:util";
 import { resolve } from "node:path";
 import { runAudit } from "./eval.js";
 import { runFix } from "./fix.js";
+import { runContextAudit } from "./context.js";
+import { runDoctor } from "./doctor.js";
 import { listSkills, generateCases, genConfig } from "./gen.js";
 import { loadConfig, parseProbability, parseIntFlag, ConfigError } from "./config.js";
-import { renderTable, renderMarkdown, renderJson, renderFix } from "./report.js";
+import { renderTable, renderMarkdown, renderJson, renderFix, renderContext, renderDoctor } from "./report.js";
 const HELP = `skill-probe — audit a co-loaded agent skill library by real activation behavior.
 
 USAGE
   skill-probe --config <file.json> [options]                 # audit
+  skill-probe context --config <file.json> [options]         # isolation vs co-loaded activation rates
+  skill-probe doctor [--cwd <dir>] [--config <file.json>]    # preflight: setup/auth/config sanity
   skill-probe gen [--cwd <dir>] > probe.config.json          # draft a test config from your skills
   skill-probe fix --config <file.json> --skill <name> [opts] # rewrite a description, prove the lift
 
@@ -30,6 +34,10 @@ OPTIONS
       --markdown           emit a Markdown table (paste into a PR/README)
       --no-cost            hide the cost line (e.g. when on a Claude subscription)
       --quiet              suppress the live progress line on stderr
+  doctor only:
+      --cwd <dir>          project dir containing .claude/skills/ (default: .)
+      --config <file>      also sanity-check a config (expected skills exist, k/threshold feasible)
+      --skip-probe         don't run the harmless live probe (skip the auth check; no cost)
   gen only (uses ANTHROPIC_API_KEY if set, else the logged-in 'claude' CLI):
       --cwd <dir>          project dir containing .claude/skills/ (default: .)
       --per-skill <n>      should-fire prompts to draft per skill (default 3)
@@ -59,6 +67,17 @@ function clearProgress() {
         progressLen = 0;
     }
 }
+function makeContextProgress(values) {
+    if (values["json"] || values["quiet"] || !process.stderr.isTTY)
+        return undefined;
+    return (e) => {
+        const p = e.prompt.length > 28 ? e.prompt.slice(0, 25) + "…" : e.prompt;
+        const line = `  [${e.caseIndex}/${e.caseTotal}] ${e.condition.padEnd(9)} "${p}" — run ${e.validN}/${e.maxK}`;
+        const pad = Math.max(0, progressLen - line.length);
+        process.stderr.write("\r" + line + " ".repeat(pad));
+        progressLen = line.length;
+    };
+}
 async function audit(values) {
     const cfg = loadConfig(values["config"], {
         runtime: values["runtime"], model: values["model"],
@@ -74,6 +93,29 @@ async function audit(values) {
     console.log(text);
     return result.exitCode;
 }
+async function doctor(values) {
+    const result = await runDoctor({
+        cwd: values["cwd"] ?? ".",
+        ...(values["config"] ? { configPath: values["config"] } : {}),
+        ...(values["runtime"] ? { runtime: values["runtime"] } : {}),
+        ...(values["model"] ? { model: values["model"] } : {}),
+        ...(values["skip-probe"] ? { skipProbe: true } : {}),
+    });
+    console.log(values["json"] ? renderJson(result) : renderDoctor(result));
+    return result.exitCode;
+}
+async function context(values) {
+    const cfg = loadConfig(values["config"], {
+        runtime: values["runtime"], model: values["model"],
+        k: values["k"], threshold: values["threshold"],
+        conf: values["conf"],
+    });
+    const result = await runContextAudit(cfg, makeContextProgress(values));
+    clearProgress();
+    const showCost = !values["no-cost"];
+    console.log(values["json"] ? renderJson(result) : renderContext(result, { showCost }));
+    return result.exitCode;
+}
 async function gen(values) {
     const cwd = values["cwd"] ?? ".";
     const perSkill = parseIntFlag(values["per-skill"], "--per-skill", 3, 1);
@@ -122,6 +164,7 @@ async function main() {
             cwd: { type: "string" },
             "per-skill": { type: "string" },
             decoys: { type: "string" },
+            "skip-probe": { type: "boolean", default: false },
             json: { type: "boolean", default: false },
             markdown: { type: "boolean", default: false },
             "no-cost": { type: "boolean", default: false },
@@ -138,12 +181,17 @@ async function main() {
     if (cmd === "gen") {
         code = await gen(values);
     }
+    else if (cmd === "doctor") {
+        code = await doctor(values); // config is optional for doctor
+    }
     else {
         if (!values.config) {
             console.log(HELP);
             process.exit(2);
         }
-        code = cmd === "fix" ? await fix(values) : await audit(values);
+        code = cmd === "fix" ? await fix(values)
+            : cmd === "context" ? await context(values)
+                : await audit(values);
     }
     process.exit(code);
 }

package/dist/context.d.ts

@@ -0,0 +1,48 @@
+import type { Config } from "./types.ts";
+import { reliability } from "./stats.ts";
+import { type RunStats } from "./orchestrator.ts";
+/** Build a throwaway project dir whose .claude/skills/ holds ONLY the named skills (copied from
+ * srcCwd). Returns its path plus a cleanup(). The probe runs against this dir to load a subset. */
+export declare function buildContextDir(srcCwd: string, skillNames: string[]): {
+    cwd: string;
+    cleanup: () => void;
+};
+type Rel = ReturnType<typeof reliability>;
+interface Condition {
+    stats: RunStats;
+    rel: Rel;
+}
+export interface ContextCaseResult {
+    prompt: string;
+    expected: string;
+    isolation: Condition;
+    coLoaded: Condition;
+    deltaP: number;
+    fisherP: number;
+    /** infra failures dominated one side — the comparison is not trustworthy (NOT a behavioral result) */
+    untrustworthy: boolean;
+    /** fires reliably alone but drops significantly when the full library is co-loaded */
+    interference: boolean;
+}
+export interface ContextAuditResult {
+    runtime: string;
+    model: string;
+    threshold: number;
+    conf: number;
+    librarySize: number;
+    cases: ContextCaseResult[];
+    skipped: string[];
+    totalCost: number;
+    /** 0 = no interference; 1 = a skill is suppressed under load; 2 = an untrustworthy (infra) case */
+    exitCode: 0 | 1 | 2;
+}
+export interface ContextProgress {
+    caseIndex: number;
+    caseTotal: number;
+    prompt: string;
+    condition: "isolation" | "co-loaded";
+    validN: number;
+    maxK: number;
+}
+export declare function runContextAudit(cfg: Config, onProgress?: (e: ContextProgress) => void): Promise<ContextAuditResult>;
+export {};

package/dist/context.js

@@ -0,0 +1,112 @@
+/** `skill-probe context` — activation-rate-BY-CONTEXT. The audit measures every skill co-loaded
+ * (the real, hard condition). This command additionally measures each skill in ISOLATION (only that
+ * skill present) and compares the two: it catches the "fires fine alone, fails under load" case that
+ * a single aggregate rate hides. The drop is tested with Fisher's exact test, so interference is
+ * reported as a real effect (p-value), not eyeballed. */
+import { mkdtempSync, mkdirSync, cpSync, rmSync, existsSync } from "node:fs";
+import { tmpdir } from "node:os";
+import { join } from "node:path";
+import { reliability, fisherExact2x2 } from "./stats.js";
+import { measure } from "./orchestrator.js";
+import { getAdapter } from "./adapters/index.js";
+import { listSkills } from "./gen.js";
+import { ConfigError } from "./config.js";
+/** Same rule the audit uses (eval.ts): a measurement is untrustworthy when infrastructure
+ * failures dominate — no valid probes, or errors at least equal the valid count. Such a
+ * condition must NOT be read as a behavioral result (a runtime outage is not a clean pass). */
+function infraUntrustworthy(s) {
+    return s.n === 0 || (s.errors > 0 && s.errors >= s.n);
+}
+/** Build a throwaway project dir whose .claude/skills/ holds ONLY the named skills (copied from
+ * srcCwd). Returns its path plus a cleanup(). The probe runs against this dir to load a subset. */
+export function buildContextDir(srcCwd, skillNames) {
+    const root = mkdtempSync(join(tmpdir(), "skill-probe-ctx-"));
+    const destSkills = join(root, ".claude", "skills");
+    mkdirSync(destSkills, { recursive: true });
+    for (const name of skillNames) {
+        const src = join(srcCwd, ".claude", "skills", name);
+        if (!existsSync(src))
+            throw new Error(`skill not found in ${srcCwd}: ${name}`);
+        cpSync(src, join(destSkills, name), { recursive: true });
+    }
+    return { cwd: root, cleanup: () => rmSync(root, { recursive: true, force: true }) };
+}
+export async function runContextAudit(cfg, onProgress) {
+    const adapter = getAdapter(cfg.runtime);
+    const library = listSkills(cfg.cwd);
+    const present = new Set(library.map((s) => s.name));
+    const measured = [];
+    const skipped = [];
+    let totalCost = 0;
+    // decoys (null expected) have nothing to isolate → skipped. An expected skill that ISN'T in the
+    // library is almost always a config typo: fail loud BEFORE spending probes, never silently skip.
+    const eligible = [];
+    const unknown = [];
+    for (const c of cfg.cases) {
+        if (c.expected === null)
+            skipped.push(c.prompt);
+        else if (!present.has(c.expected))
+            unknown.push(c.expected);
+        else
+            eligible.push(c);
+    }
+    if (unknown.length) {
+        throw new ConfigError([
+            `expected skill(s) not in ${cfg.cwd}/.claude/skills/: ${[...new Set(unknown)].join(", ")} ` +
+                `(typo?). Available: ${library.map((s) => s.name).join(", ") || "(none)"}.`,
+        ]);
+    }
+    const common = { maxK: cfg.k, threshold: cfg.threshold, conf: cfg.conf, ...(cfg.model ? { model: cfg.model } : {}) };
+    for (let i = 0; i < eligible.length; i++) {
+        const c = eligible[i];
+        const expected = c.expected;
+        // isolation: a temp project containing only this skill
+        const ctx = buildContextDir(cfg.cwd, [expected]);
+        let iso;
+        try {
+            iso = await measure(adapter, c.prompt, expected, {
+                cwd: ctx.cwd, ...common,
+                ...(onProgress ? { onProbe: (validN, maxK) => onProgress({ caseIndex: i + 1, caseTotal: eligible.length, prompt: c.prompt, condition: "isolation", validN, maxK }) } : {}),
+            });
+        }
+        finally {
+            ctx.cleanup();
+        }
+        // co-loaded: the real library at cfg.cwd
+        const co = await measure(adapter, c.prompt, expected, {
+            cwd: cfg.cwd, ...common,
+            ...(onProgress ? { onProbe: (validN, maxK) => onProgress({ caseIndex: i + 1, caseTotal: eligible.length, prompt: c.prompt, condition: "co-loaded", validN, maxK }) } : {}),
+        });
+        totalCost += iso.totalCost + co.totalCost;
+        const isoRel = reliability(iso.hits, iso.n, cfg.conf);
+        const coRel = reliability(co.hits, co.n, cfg.conf);
+        // 2x2: rows = isolation/co-loaded, cols = hit/miss
+        const fisherP = (iso.n > 0 && co.n > 0)
+            ? fisherExact2x2(iso.hits, iso.n - iso.hits, co.hits, co.n - co.hits)
+            : 1;
+        const deltaP = coRel.pHat - isoRel.pHat;
+        // if infra failures dominated either side, the comparison isn't a behavioral result at all —
+        // it must never read as "no interference / pass" just because fisherP defaulted to 1.
+        const untrustworthy = infraUntrustworthy(iso) || infraUntrustworthy(co);
+        const interference = !untrustworthy &&
+            isoRel.pHat >= cfg.threshold && // fires reliably alone
+            deltaP < 0 && // and drops when co-loaded
+            fisherP < 0.05; // and the drop is statistically real
+        measured.push({
+            prompt: c.prompt, expected,
+            isolation: { stats: iso, rel: isoRel },
+            coLoaded: { stats: co, rel: coRel },
+            deltaP, fisherP, untrustworthy, interference,
+        });
+    }
+    // precedence mirrors the audit: a real behavioral signal (interference, exit 1) outranks an
+    // untrustworthy/infra case (exit 2), which outranks a clean run (exit 0).
+    const exitCode = measured.some((m) => m.interference) ? 1
+        : measured.some((m) => m.untrustworthy) ? 2
+            : 0;
+    return {
+        runtime: cfg.runtime, model: cfg.model ?? "(runtime default)",
+        threshold: cfg.threshold, conf: cfg.conf, librarySize: library.length,
+        cases: measured, skipped, totalCost, exitCode,
+    };
+}

package/dist/doctor.d.ts

@@ -0,0 +1,23 @@
+import type { ProbeResult, ProbeOpts } from "./types.ts";
+export type CheckStatus = "pass" | "warn" | "fail" | "info";
+export interface Check {
+    status: CheckStatus;
+    message: string;
+}
+export interface DoctorResult {
+    checks: Check[];
+    exitCode: 0 | 1 | 2;
+}
+export interface DoctorOpts {
+    cwd: string;
+    configPath?: string;
+    runtime?: string;
+    model?: string;
+    skipProbe?: boolean;
+}
+/** Injection seam so tests don't shell out to a real runtime. */
+export interface DoctorDeps {
+    cliInstalled?: (cmd: string) => boolean;
+    probe?: (runtime: string, prompt: string, opts: ProbeOpts) => Promise<ProbeResult>;
+}
+export declare function runDoctor(opts: DoctorOpts, deps?: DoctorDeps): Promise<DoctorResult>;

package/dist/doctor.js

@@ -0,0 +1,138 @@
+/** `skill-probe doctor` — preflight checks. Now that the tool is public, the first failures users
+ * hit are SETUP failures (no skills dir, runtime not installed, not authed, a config typo), not deep
+ * eval problems. doctor surfaces those in one cheap pass before they spend probes on a broken setup.
+ *
+ * Exit: 0 = healthy, 1 = warnings only, 2 = at least one hard failure. */
+import { spawnSync } from "node:child_process";
+import { existsSync, readdirSync, readFileSync } from "node:fs";
+import { join, resolve } from "node:path";
+import { loadConfig } from "./config.js";
+import { wilsonInterval } from "./stats.js";
+import { kToCertify } from "./report.js";
+import { getAdapter } from "./adapters/index.js";
+const NAME_RE = /^name:\s*(.*)$/m;
+const DESC_RE = /^description:\s*(.*)$/m;
+const MIN_NODE = { major: 22, minor: 18 }; // package.json engines: ">=22.18"
+function realCliInstalled(cmd) {
+    const r = spawnSync(cmd, ["--version"], { stdio: "ignore", timeout: 15_000 });
+    // ENOENT = not on PATH. Any other result (even a non-zero exit) means the binary exists.
+    return !(r.error && r.error.code === "ENOENT");
+}
+export async function runDoctor(opts, deps = {}) {
+    const cliInstalled = deps.cliInstalled ?? realCliInstalled;
+    const probe = deps.probe ?? ((rt, p, o) => getAdapter(rt).probe(p, o));
+    const checks = [];
+    const add = (status, message) => { checks.push({ status, message }); };
+    // 1. Node version (declared engine)
+    const [maj = 0, min = 0] = process.versions.node.split(".").map(Number);
+    if (maj > MIN_NODE.major || (maj === MIN_NODE.major && min >= MIN_NODE.minor)) {
+        add("pass", `Node ${process.versions.node}`);
+    }
+    else {
+        add("warn", `Node ${process.versions.node} is below the declared engine >=22.18 ` +
+            `(the compiled CLI may still run, but this is untested)`);
+    }
+    // 2. Config (optional). Load first — it can redirect the skills dir (cwd resolves to the config's dir).
+    let cfg = null;
+    if (opts.configPath) {
+        if (!existsSync(resolve(opts.configPath))) {
+            add("fail", `config not found: ${opts.configPath}`);
+        }
+        else {
+            try {
+                cfg = loadConfig(opts.configPath, { ...(opts.runtime ? { runtime: opts.runtime } : {}) });
+                add("pass", `config parsed: ${opts.configPath}`);
+            }
+            catch (e) {
+                add("fail", `config ${opts.configPath}: ${e instanceof Error ? e.message : String(e)}`);
+            }
+        }
+    }
+    const skillsCwd = cfg ? cfg.cwd : resolve(opts.cwd);
+    const runtime = cfg?.runtime ?? opts.runtime ?? "claude-code";
+    // 3-6. Skills directory + each SKILL.md
+    const skillNames = new Set();
+    const skillsRoot = join(skillsCwd, ".claude", "skills");
+    if (!existsSync(skillsRoot)) {
+        add("fail", `no .claude/skills/ directory under ${skillsCwd}`);
+    }
+    else {
+        const dirs = readdirSync(skillsRoot, { withFileTypes: true }).filter((d) => d.isDirectory());
+        if (dirs.length === 0) {
+            add("fail", `.claude/skills/ exists but contains no skill folders`);
+        }
+        let valid = 0;
+        for (const d of dirs) {
+            const md = join(skillsRoot, d.name, "SKILL.md");
+            if (!existsSync(md)) {
+                add("fail", `skill "${d.name}" missing SKILL.md`);
+                continue;
+            }
+            const text = readFileSync(md, "utf8");
+            const name = NAME_RE.exec(text)?.[1]?.trim() ?? "";
+            const desc = DESC_RE.exec(text)?.[1]?.trim() ?? "";
+            if (!name)
+                add("fail", `skill "${d.name}" missing name:`);
+            else if (name !== d.name)
+                add("warn", `skill "${d.name}" has name: "${name}" (doesn't match folder name)`);
+            if (!desc)
+                add("fail", `skill "${d.name}" missing description:`);
+            else if (desc === "|" || desc === ">")
+                add("warn", `skill "${d.name}" uses a multi-line description block — a single-line description triggers more reliably`);
+            skillNames.add(d.name);
+            valid++;
+        }
+        if (valid > 0)
+            add("pass", `found .claude/skills/ with ${valid} skill${valid === 1 ? "" : "s"}`);
+    }
+    // 7-9. Config-dependent checks
+    if (cfg) {
+        const missing = cfg.cases
+            .map((c) => c.expected).filter((s) => s !== null)
+            .filter((s) => !skillNames.has(s));
+        if (missing.length)
+            add("fail", `config expects skill(s) not in the library: ${[...new Set(missing)].join(", ")}`);
+        else
+            add("pass", `expected skills all exist`);
+        if (cfg.cases.every((c) => c.expected === null)) {
+            add("warn", `config has only decoy cases — no skill is being verified to fire`);
+        }
+        // a strict threshold at a small k can be mathematically impossible to certify
+        if (wilsonInterval(cfg.k, cfg.k, cfg.conf).lo <= cfg.threshold) {
+            add("warn", `threshold ${Math.round(cfg.threshold * 100)}% with k=${cfg.k} cannot certify a pass ` +
+                `even on a perfect run — need ~k=${kToCertify(cfg.threshold, cfg.conf)} (or lower the threshold)`);
+        }
+    }
+    // 10-11. Runtime + a harmless live probe (verifies the CLI is installed AND authed)
+    add("info", `runtime: ${runtime}  model: ${cfg?.model ?? opts.model ?? "(runtime default)"}`);
+    const known = runtime === "claude-code" || runtime === "opencode";
+    const bin = runtime === "claude-code" ? "claude" : runtime === "opencode" ? "opencode" : null;
+    let cliOk = false;
+    if (!known) {
+        add("fail", `unknown runtime "${runtime}" (expected claude-code or opencode)`);
+    }
+    else if (bin && cliInstalled(bin)) {
+        add("pass", `${bin} CLI found`);
+        cliOk = true;
+    }
+    else {
+        add("fail", `${bin} CLI not found on PATH — install it and authenticate to run ${runtime}`);
+    }
+    if (opts.skipProbe) {
+        add("info", `probe skipped (--skip-probe)`);
+    }
+    else if (known && cliOk) {
+        const r = await probe(runtime, "Reply with exactly: ok", {
+            cwd: skillsCwd, timeoutMs: 60_000,
+            ...(cfg?.model ?? opts.model ? { model: (cfg?.model ?? opts.model) } : {}),
+        });
+        if (r.status === "ok")
+            add("pass", `${runtime} probe succeeded (CLI is authenticated)`);
+        else
+            add("fail", `${runtime} probe failed — check auth/login: ${r.error ?? "unknown error"}`);
+    }
+    const hasFail = checks.some((c) => c.status === "fail");
+    const hasWarn = checks.some((c) => c.status === "warn");
+    const exitCode = hasFail ? 2 : hasWarn ? 1 : 0;
+    return { checks, exitCode };
+}

package/dist/report.d.ts

@@ -1,5 +1,9 @@
 import type { AuditResult } from "./eval.ts";
 import type { FixResult } from "./fix.ts";
+import type { ContextAuditResult } from "./context.ts";
+import type { DoctorResult } from "./doctor.ts";
+/** Smallest k at which an all-success run could certify reliability >= threshold. */
+export declare function kToCertify(threshold: number, conf: number): number;
 export interface RenderOpts {
     showCost?: boolean;
 }
@@ -7,4 +11,7 @@ export declare function renderTable(a: AuditResult, opts?: RenderOpts): string;
 /** Markdown table — paste into a PR or README. */
 export declare function renderMarkdown(a: AuditResult, opts?: RenderOpts): string;
 export declare function renderJson(a: AuditResult | unknown): string;
+export declare function renderDoctor(d: DoctorResult): string;
+/** activation-rate-by-context: isolation vs co-loaded, with a Fisher's-exact p on the drop. */
+export declare function renderContext(a: ContextAuditResult, opts?: RenderOpts): string;
 export declare function renderFix(f: FixResult): string;

package/dist/report.js

@@ -2,12 +2,13 @@
  * errors are shown distinctly from behavioral results. */
 import { wilsonInterval } from "./stats.js";
 const pct = (x) => `${Math.round(x * 100)}%`;
+const pval = (p) => (p < 0.001 ? "<0.001" : p.toFixed(3));
 const LABEL = {
     pass: "PASS        ", fail: "FAIL        ",
     inconclusive: "INCONCLUSIVE", error: "ERROR       ",
 };
 /** Smallest k at which an all-success run could certify reliability >= threshold. */
-function kToCertify(threshold, conf) {
+export function kToCertify(threshold, conf) {
     for (let k = 1; k <= 500; k++) {
         if (wilsonInterval(k, k, conf).lo > threshold)
             return k;
@@ -80,6 +81,62 @@ export function renderMarkdown(a, opts = {}) {
 export function renderJson(a) {
     return JSON.stringify(a, null, 2);
 }
+export function renderDoctor(d) {
+    const out = ["skill-probe doctor", ""];
+    for (const c of d.checks)
+        out.push(`${c.status.toUpperCase().padEnd(4)}  ${c.message}`);
+    out.push("");
+    const f = d.checks.filter((c) => c.status === "fail").length;
+    const w = d.checks.filter((c) => c.status === "warn").length;
+    out.push(f ? `${f} failure(s)` + (w ? `, ${w} warning(s)` : "") + `  |  exit ${d.exitCode}`
+        : w ? `${w} warning(s)  |  exit ${d.exitCode}`
+            : `all healthy  |  exit 0`);
+    return out.join("\n");
+}
+/** activation-rate-by-context: isolation vs co-loaded, with a Fisher's-exact p on the drop. */
+export function renderContext(a, opts = {}) {
+    const showCost = opts.showCost ?? true;
+    const out = [];
+    out.push(`skill-probe context — runtime: ${a.runtime}  model: ${a.model}  ` +
+        `threshold: ${pct(a.threshold)}  library: ${a.librarySize} skills`);
+    out.push(`  isolation = only that skill loaded · co-loaded = the full library (${a.librarySize}) loaded`);
+    out.push("");
+    const ci = (r) => `${pct(r.pHat)} [${pct(r.ciLow)}, ${pct(r.ciHigh)}] k=${r.k}`;
+    for (const c of a.cases) {
+        if (c.untrustworthy) {
+            out.push(`  [⛔ ERROR]  ${c.expected}  | ${c.prompt}`);
+            out.push(`        infrastructure errors dominated — comparison NOT trustworthy ` +
+                `(isolation ${c.isolation.stats.errors} err/${c.isolation.stats.n} valid, ` +
+                `co-loaded ${c.coLoaded.stats.errors} err/${c.coLoaded.stats.n} valid)`);
+            const reason = c.isolation.stats.lastError ?? c.coLoaded.stats.lastError;
+            if (reason)
+                out.push(`        reason: ${reason}`);
+            continue;
+        }
+        const flag = c.interference ? "⚠ INTERFERENCE" : "ok";
+        out.push(`  [${flag}]  ${c.expected}  | ${c.prompt}`);
+        out.push(`        isolation ${ci(c.isolation.rel)}`);
+        out.push(`        co-loaded ${ci(c.coLoaded.rel)}`);
+        out.push(`        Δ ${c.deltaP >= 0 ? "+" : ""}${pct(c.deltaP)} under load   Fisher p=${pval(c.fisherP)}`);
+        if (c.interference) {
+            const thieves = Object.keys(c.coLoaded.stats.dist)
+                .filter((s) => s !== c.expected && s !== "None");
+            out.push(`        ↳ fires reliably alone but is suppressed when the library is co-loaded` +
+                (thieves.length ? ` (stolen by: ${thieves.join(", ")})` : ""));
+        }
+    }
+    if (a.skipped.length) {
+        out.push("");
+        out.push(`  skipped ${a.skipped.length} decoy case(s) — no single skill to isolate.`);
+    }
+    out.push("");
+    const interfN = a.cases.filter((c) => c.interference).length;
+    const errN = a.cases.filter((c) => c.untrustworthy).length;
+    out.push(`Result: ${interfN} interference / ${a.cases.length} measured` +
+        (errN ? ` / ${errN} error` : "") + `  |  exit ${a.exitCode}` +
+        (showCost ? `  |  cost $${a.totalCost.toFixed(4)}` : ""));
+    return out.join("\n");
+}
 export function renderFix(f) {
     const out = [];
     out.push(`skill-probe fix — skill: ${f.skill}`);

package/package.json

@@ -1,9 +1,11 @@
 {
   "name": "skill-probe",
-  "version": "0.4.0",
+  "version": "0.6.0",
   "description": "Runtime-agnostic, co-loaded-aware auditor for AI agent skill libraries: measures which skill actually fires, with statistical confidence.",
   "type": "module",
-  "bin": { "skill-probe": "dist/cli.js" },
+  "bin": {
+    "skill-probe": "dist/cli.js"
+  },
   "scripts": {
     "dev": "node src/cli.ts",
     "test": "node --test test/*.test.ts",
@@ -11,13 +13,36 @@
     "build": "tsc",
     "prepublishOnly": "npm run build && npm test"
   },
-  "keywords": ["agent-skills", "skill", "eval", "claude-code", "mcp", "SKILL.md", "trigger"],
+  "keywords": [
+    "agent-skills",
+    "skill",
+    "eval",
+    "claude-code",
+    "mcp",
+    "SKILL.md",
+    "trigger"
+  ],
   "author": "HystonKayange",
   "license": "MIT",
-  "repository": { "type": "git", "url": "git+https://github.com/HystonKayange/skill-probe.git" },
+  "repository": {
+    "type": "git",
+    "url": "git+https://github.com/HystonKayange/skill-probe.git"
+  },
   "homepage": "https://github.com/HystonKayange/skill-probe#readme",
-  "bugs": { "url": "https://github.com/HystonKayange/skill-probe/issues" },
-  "engines": { "node": ">=22.18" },
-  "devDependencies": { "typescript": "^5.5.0", "@types/node": "^22.0.0" },
-  "files": ["dist", "examples", "README.md", "LICENSE"]
+  "bugs": {
+    "url": "https://github.com/HystonKayange/skill-probe/issues"
+  },
+  "engines": {
+    "node": ">=22.18"
+  },
+  "devDependencies": {
+    "typescript": "^5.5.0",
+    "@types/node": "^22.0.0"
+  },
+  "files": [
+    "dist",
+    "examples",
+    "README.md",
+    "LICENSE"
+  ]
 }