gm-skill 2.0.2571 → 2.0.2573

package/AGENTS.md

@@ -172,6 +172,16 @@ A task that reduces to read/investigate/report, or a change confined to files th
 
 **Bootstrap contract**: `ensureReady` inits wasm hook-free, sha256-rewrites stale installed SKILL.md, seeds per-project `CLAUDE.md`/`.gm/next-step.md`. Detail: the recall store (`recall: skill-initiated bootstrap contract`, `recall: SKILL.md auto-refresh`).
 
+## Supply-chain tell scanner
+
+`scripts/scan-supply-chain-tells.mjs` scans any file tree for the injected-backdoor/AI-tell signature classes found in a real incident on a consumer project (freddie): an obfuscated C2 stager appended after a legitimate `export default` in a build config, delivered via a compromised automated release commit that also bumped a dependency and dropped `.env` from `.gitignore`. Run it: `node scripts/scan-supply-chain-tells.mjs <path...>` (defaults to cwd). Exit 0 = clean, 1 = findings, 2 = scan error.
+
+Detection classes: exact-match known-incident strings (campaign tag, hardcoded C2-lookup address, custom payload-smuggling header), structural patterns (blockchain-RPC-derived C2 address resolution, IP-from-bytes decode, detached+hidden-stdio child-process spawn, `eval()` fed by a network response, XOR-decode loops, `require`/`module` stashed onto `global`, a dense minified tail line appended to an otherwise normal readable file, `\uXXXX`-escaped ASCII used to defeat plaintext grep, multiple public RPC endpoints raced with `Promise.any` for C2 resilience), and Unicode confusables/invisible characters (zero-width spaces/joiners, bidi override characters — the Trojan Source class — soft hyphens, and common Cyrillic homoglyphs for Latin letters).
+
+**Dispatch this scan at EMIT (before any live witness) and again at DECIDE's adversarial corner-case sweep, on every file the diff touches** — alongside the existing empty/overflow/injection/resource-exhaustion sweep classes, not instead of them. A `security` toolset dependency bump, an automated release commit, or any file with unexpectedly dense/obfuscated content are the highest-suspicion triggers for an out-of-band run against the full tree, not just the diff. A finding routes like any other DECIDE discovery: confirmed-malicious content is a `transition to=EMIT` fix (remove and re-witness), never a silent ignore or a "looks fine" without running the scan.
+
+This is a heuristic signature scanner, not a guarantee — it catches the specific technique classes seen in the confirmed incident and known-adjacent variants; it does not replace `npm audit`, dependency pinning review, or reading an unfamiliar dependency's actual source before trusting an automated bump.
+
 ## Cascade pipeline
 
 Push to any rs-* sibling -> `cascade.yml` -> rs-plugkit `release.yml` -> single `plugkit.wasm` (npm `plugkit-wasm` + `plugkit-bin` Releases) -> best-effort gm-metadata-sync commit -> `publish.yml` ships gm-skill+gm-plugkit+SKILL.md mirror. Step sequence + PUBLISHER_TOKEN: the recall store (`recall: cascade pipeline`).

package/gm-plugkit/package.json

@@ -1,6 +1,6 @@
 {
   "name": "gm-plugkit",
-  "version": "2.0.2571",
+  "version": "2.0.2573",
   "description": "Bootstrap and daemon-spawn tool for gm plugkit binary. Downloads the correct platform wasm, verifies SHA256, and launches agentplug-runner (the native wasm host) as the spool watcher daemon.",
   "main": "index.js",
   "bin": {

package/gm.json

@@ -1,6 +1,6 @@
 {
   "name": "gm",
-  "version": "2.0.2571",
+  "version": "2.0.2573",
   "description": "Spool-dispatch orchestration engine with unified state machine, skills, and automated git enforcement",
   "author": "AnEntrypoint",
   "license": "MIT",

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "gm-skill",
-  "version": "2.0.2571",
+  "version": "2.0.2573",
   "description": "Canonical universal harness — AI-native software engineering via skill-driven orchestration; bootstraps plugkit for task execution and session isolation. Install in any AI coding agent host.",
   "author": "AnEntrypoint",
   "license": "MIT",

package/scripts/scan-supply-chain-tells.mjs

@@ -0,0 +1,257 @@
+#!/usr/bin/env node
+// Scans files for supply-chain-backdoor and AI-injection "tells" -- signatures
+// found live in a real incident (an obfuscated C2 stager appended to
+// vite.browser.config.js, delivered via a compromised automated release
+// commit that also bumped a dependency and dropped .env from .gitignore).
+//
+// Run standalone: node scripts/scan-supply-chain-tells.mjs [path...]
+// Exit code 0 = clean, 1 = findings, 2 = scan error.
+//
+// Designed to be invoked from any project (not just this one) -- pass the
+// target repo root(s) as argv, or it scans cwd.
+
+import fs from 'node:fs'
+import path from 'node:path'
+import process from 'node:process'
+
+const SKIP_DIRS = new Set([
+  'node_modules', '.git', 'dist', 'build', '.next', 'vendor', '.cache',
+  '.svelte-kit', '.nuxt', '.output', '.turbo', 'out', 'coverage', '.parcel-cache',
+])
+// Code files only. JSON/YAML/MD routinely carry legitimate non-Latin natural-
+// language text (Cyrillic, Greek, CJK, etc.) which is indistinguishable from a
+// homoglyph attack by codepoint alone -- the Unicode-confusable check below is
+// only meaningful applied to CODE, where an identifier/URL is expected to be
+// plain ASCII and non-ASCII inside one is a genuine anomaly, not content.
+const CODE_EXT = new Set(['.js', '.mjs', '.cjs', '.ts', '.tsx', '.jsx', '.sh', '.ps1', '.py'])
+// Broader set for the exact-string / pattern checks, which key on code shapes
+// (require calls, eval, spawn) that can legitimately only appear in code or
+// config, not prose -- config/build files are still worth the exact-string and
+// structural-pattern passes, just not the Unicode-confusable pass.
+const TEXT_EXT = new Set([...CODE_EXT, '.json', '.yml', '.yaml', '.md'])
+
+// ---- Exact-match signatures from the confirmed incident ----------------
+const EXACT_STRINGS = [
+  { sig: 'A9-2057', why: 'campaign/version tag literal seen in a live C2 stager' },
+  { sig: '0xa322E5f3D311D3080e6f0121063e9aDC2490Ef1a', why: 'hardcoded Ethereum address used as a blockchain-based C2 config lookup key' },
+  { sig: 'eth.blockscout.com/api', why: 'block-explorer API used as a C2-resolution fallback' },
+  { sig: 'x-payload-b64', why: 'custom HTTP header used to smuggle a staged payload' },
+]
+
+// ---- Structural / behavioral patterns (regex) ---------------------------
+const PATTERNS = [
+  {
+    name: 'blockchain-derived-c2-lookup',
+    re: /eth_getBlockByNumber|eth_getTransactionCount|eth_blockNumber/,
+    why: 'blockchain RPC calls used to derive a C2 address/config are a known dead-drop technique — legitimate web3 code should be an isolated, obviously-named module, not appended to an unrelated build config',
+  },
+  {
+    name: 'ip-from-bytes-decode',
+    re: /\[0\]\s*\+\s*['"]\.['"]\s*\+\s*\w+\[1\]\s*\+\s*['"]\.['"]\s*\+\s*\w+\[2\]/,
+    why: 'byte-array-to-dotted-IP decode shape, commonly used to hide a C2 address inside binary data (a tx field, an image, etc.)',
+  },
+  {
+    name: 'detached-hidden-spawn',
+    re: /spawn\s*\(\s*['"]node['"][\s\S]{0,120}detached\s*:\s*true[\s\S]{0,120}stdio\s*:\s*['"]ignore['"]/,
+    why: 'detached + stdio:ignore + unref() child process launch — runs code that outlives and is invisible to the parent process',
+  },
+  {
+    name: 'detached-hidden-spawn-loose',
+    re: /detached\s*:\s*true[\s\S]{0,60}(stdio\s*:\s*['"]ignore['"]|windowsHide\s*:\s*true)/,
+    why: 'detached background process with hidden stdio/window — legitimate daemonization exists but is rare in application/build code and should be named and commented',
+  },
+  {
+    name: 'eval-of-network-response',
+    re: /eval\s*\(\s*[\w.]*\+?\s*(await\s+)?\w*(fetch|http|https|response|body|payload|res)\w*/i,
+    why: 'eval() fed directly by network response content — remote code execution primitive',
+  },
+  {
+    name: 'xor-decode-helper',
+    re: /charCodeAt\(\w+\s*%\s*\w+\.length\)/,
+    why: 'per-byte XOR-against-key loop, the standard shape for decoding an obfuscated payload at runtime',
+  },
+  {
+    name: 'global-require-module-stash',
+    re: /global\.\w+\s*=\s*require\s*;\s*typeof\s+module\s*===\s*['"]object['"]/,
+    why: 'stashing require/module onto global — used so eval()-ed code can access them outside its lexical scope',
+  },
+  {
+    name: 'minified-tail-after-normal-code',
+    re: /\n[a-z]\.[a-z]\([a-z],[a-z],[a-z]\)/,
+    why: 'single-letter-identifier minified code appearing after normally-formatted source — mixing styles in one file is a strong injection tell',
+    weak: true,
+  },
+  {
+    name: 'promise-any-rpc-race',
+    re: /Promise\.any\(\s*\[?\s*['"]https?:\/\/[^,\]]+,\s*['"]https?:\/\//,
+    why: 'racing multiple public RPC/API endpoints for resilience — common in C2 code designed to survive one endpoint being blocked',
+  },
+]
+
+// ---- Unicode confusables / invisible-character tells ---------------------
+// Each entry: a *visible-as-ASCII-but-isn't* or fully invisible codepoint.
+const SUSPICIOUS_UNICODE = [
+  { cp: 0x200b, name: 'ZERO WIDTH SPACE' },
+  { cp: 0x200c, name: 'ZERO WIDTH NON-JOINER' },
+  { cp: 0x200d, name: 'ZERO WIDTH JOINER' },
+  { cp: 0x200e, name: 'LEFT-TO-RIGHT MARK' },
+  { cp: 0x200f, name: 'RIGHT-TO-LEFT MARK' },
+  { cp: 0x202a, name: 'LEFT-TO-RIGHT EMBEDDING' },
+  { cp: 0x202b, name: 'RIGHT-TO-LEFT EMBEDDING' },
+  { cp: 0x202c, name: 'POP DIRECTIONAL FORMATTING' },
+  { cp: 0x202d, name: 'LEFT-TO-RIGHT OVERRIDE' },
+  { cp: 0x202e, name: 'RIGHT-TO-LEFT OVERRIDE (Trojan Source bidi attack)' },
+  { cp: 0x2060, name: 'WORD JOINER' },
+  { cp: 0x2066, name: 'LEFT-TO-RIGHT ISOLATE' },
+  { cp: 0x2067, name: 'RIGHT-TO-LEFT ISOLATE' },
+  { cp: 0x2068, name: 'FIRST STRONG ISOLATE' },
+  { cp: 0x2069, name: 'POP DIRECTIONAL ISOLATE' },
+  { cp: 0xfeff, name: 'ZERO WIDTH NO-BREAK SPACE / BOM (mid-file)' },
+  { cp: 0x00ad, name: 'SOFT HYPHEN' },
+  // Common homoglyphs used to disguise identifiers/URLs (Cyrillic look-alikes)
+  { cp: 0x0410, name: 'CYRILLIC CAPITAL А (looks like Latin A)' },
+  { cp: 0x0430, name: 'CYRILLIC SMALL а (looks like Latin a)' },
+  { cp: 0x0415, name: 'CYRILLIC CAPITAL Е (looks like Latin E)' },
+  { cp: 0x0435, name: 'CYRILLIC SMALL е (looks like Latin e)' },
+  { cp: 0x041e, name: 'CYRILLIC CAPITAL О (looks like Latin O)' },
+  { cp: 0x043e, name: 'CYRILLIC SMALL о (looks like Latin o)' },
+  { cp: 0x0420, name: 'CYRILLIC CAPITAL Р (looks like Latin P)' },
+  { cp: 0x0440, name: 'CYRILLIC SMALL р (looks like Latin p)' },
+  { cp: 0x0421, name: 'CYRILLIC CAPITAL С (looks like Latin C)' },
+  { cp: 0x0441, name: 'CYRILLIC SMALL с (looks like Latin c)' },
+]
+
+// A file whose non-ASCII payload is almost entirely \uXXXX-style JS escape
+// sequences decoding to plain ASCII is itself a tell (deliberate obfuscation
+// to defeat plain-string grep, seen in the confirmed incident's http/https/
+// url/child_process require() calls).
+const ESCAPED_ASCII_RUN = /(\\u00[2-7][0-9a-fA-F]){6,}/
+
+function walk(dir, out) {
+  let entries
+  try {
+    entries = fs.readdirSync(dir, { withFileTypes: true })
+  } catch {
+    return
+  }
+  for (const e of entries) {
+    if (SKIP_DIRS.has(e.name)) continue
+    const p = path.join(dir, e.name)
+    if (e.isDirectory()) {
+      walk(p, out)
+    } else if (e.isFile() && TEXT_EXT.has(path.extname(e.name))) {
+      out.push(p)
+    }
+  }
+}
+
+function scanFile(filePath) {
+  let text
+  try {
+    text = fs.readFileSync(filePath, 'utf8')
+  } catch {
+    return []
+  }
+  const findings = []
+
+  for (const { sig, why } of EXACT_STRINGS) {
+    if (text.includes(sig)) {
+      findings.push({ filePath, kind: 'exact-string', sig, why })
+    }
+  }
+
+  for (const { name, re, why, weak } of PATTERNS) {
+    const m = text.match(re)
+    if (m) {
+      const line = text.slice(0, m.index).split('\n').length
+      findings.push({ filePath, kind: weak ? 'pattern-weak' : 'pattern', name, why, line })
+    }
+  }
+
+  const lines = text.split('\n')
+  const nonBlank = lines.filter(l => l.trim().length > 0)
+  if (nonBlank.length >= 5) {
+    const lens = nonBlank.map(l => l.length)
+    const avgLen = lens.reduce((a, b) => a + b, 0) / lens.length
+    const lastLen = lens[lens.length - 1]
+    const looksMinified = /[;,]\s*[a-zA-Z_$][\w$]*\s*=/.test(nonBlank[nonBlank.length - 1]) &&
+      /\b(function|=>|require\(|const |let |var )\b/.test(nonBlank[nonBlank.length - 1])
+    if (lastLen > 2000 && lastLen > avgLen * 20 && looksMinified) {
+      findings.push({
+        filePath,
+        kind: 'pattern',
+        name: 'dense-minified-tail-line',
+        line: nonBlank.length,
+        why: `final non-blank line is ${lastLen} chars (${Math.round(lastLen / avgLen)}x the file's average line length) and looks like minified code — this is the exact shape of a payload appended to the end of an otherwise normal, readable file`,
+      })
+    }
+  }
+
+  if (ESCAPED_ASCII_RUN.test(text)) {
+    findings.push({
+      filePath,
+      kind: 'pattern',
+      name: 'escaped-ascii-obfuscation',
+      why: '6+ consecutive \\uXXXX escapes decoding to plain ASCII — deliberate string obfuscation to defeat plaintext grep (e.g. \\u0068\\u0074\\u0074\\u0070 = "http")',
+    })
+  }
+
+  if (CODE_EXT.has(path.extname(filePath))) {
+    // De-duplicate per (codepoint, line): a run of several confusables on one
+    // line (a whole disguised word) is one finding, not one per character.
+    const seen = new Set()
+    for (let i = 0; i < text.length; i++) {
+      const cp = text.codePointAt(i)
+      const hit = SUSPICIOUS_UNICODE.find(u => u.cp === cp)
+      if (hit) {
+        const line = text.slice(0, i).split('\n').length
+        const key = cp + ':' + line
+        if (seen.has(key)) continue
+        seen.add(key)
+        findings.push({
+          filePath,
+          kind: 'unicode',
+          name: hit.name,
+          codepoint: '0x' + cp.toString(16),
+          line,
+          why: 'invisible or confusable Unicode codepoint inside a code file — used to hide code from visual review or disguise an identifier/URL as something else',
+        })
+      }
+    }
+  }
+
+  return findings
+}
+
+function main() {
+  const targets = process.argv.slice(2)
+  const roots = targets.length ? targets : [process.cwd()]
+  const files = []
+  for (const root of roots) {
+    const stat = fs.existsSync(root) ? fs.statSync(root) : null
+    if (!stat) continue
+    if (stat.isDirectory()) walk(root, files)
+    else files.push(root)
+  }
+
+  const allFindings = []
+  for (const f of files) {
+    allFindings.push(...scanFile(f))
+  }
+
+  if (!allFindings.length) {
+    console.log(`scan-supply-chain-tells: clean (${files.length} files scanned)`)
+    process.exit(0)
+  }
+
+  console.log(`scan-supply-chain-tells: ${allFindings.length} finding(s) across ${new Set(allFindings.map(f => f.filePath)).size} file(s)\n`)
+  for (const f of allFindings) {
+    const loc = f.line ? `:${f.line}` : ''
+    const label = f.sig || f.name
+    console.log(`[${f.kind}] ${f.filePath}${loc} — ${label}`)
+    console.log(`    ${f.why}`)
+  }
+  process.exit(1)
+}
+
+main()

package/skills/gm-semantic-anchors/SKILL.md

@@ -0,0 +1,262 @@
+---
+name: gm-semantic-anchors
+description: The nonlinear backreferencing graph of every named technique gm's 9-phase prose already invokes (SPECIFY through UPDATE_DOCS), cross-referenced against the llm-coding/Semantic-Anchors catalog, plus the well-known techniques that catalog is missing. Primes gm's own fsm-propose-override self-reconfiguration: a proposed prose/graph override cites the anchor it strengthens or the gap it closes, not a bare paraphrase.
+allowed-tools: Skill, Read, Write
+---
+
+# gm-semantic-anchors
+
+A semantic anchor is a named, attributed technique from well-known literature that activates an LLM's existing knowledge of that technique precisely and compactly -- "Use TDD, London School" over "write tests first, mock dependencies, work outside-in" (llm-coding/Semantic-Anchors, `what-qualifies-as-a-semantic-anchor`). gm's own phase prose already runs on this principle: each phase's "Preferences (named, narrow)" section is a curated anchor list. This skill is that list expressed as one graph, cross-checked against the public Semantic-Anchors reference (191 anchors, `github.com/llm-coding/Semantic-Anchors`), so a `fsm-propose-override` self-reconfiguration proposal can cite a precise anchor instead of composing new prose from scratch.
+
+## Why this exists
+
+`fsm-propose-override` (rs-plugkit, `orchestrator/fsm_propose.rs`) lets a session write its own `.gm/instructions/<key>.md` or `fsm/graph.json` override when `self-reconfig-candidate` fires. A proposal grounded in an attributed, well-known technique is falsifiable and reviewable; a proposal that paraphrases a vague intuition is not. This graph is the lookup table: given a phase and a friction pattern, which anchor already names the fix, and which anchors does it backreference for a fuller picture.
+
+## The graph
+
+99 anchors: 93 cross-matched by exact id against the public Semantic-Anchors catalog, 6 added because they are well-known, well-attributed techniques gm's own CONC (concurrency/performance) and RES (resilience)/STATE (correctness) phases depend on but the catalog does not yet list. Edges are the catalog's own `:related:` backreference field where fetched live from `docs/anchors/<id>.adoc`; nodes are grouped by the gm phase that names them, which is itself a valid nonlinear grouping axis (an anchor can serve a phase other than the one gm currently files it under).
+
+```mermaid
+flowchart LR
+  subgraph specify["SPECIFY"]
+    solid_principles["SOLID Principles<br/><small>Robert C. Martin</small>"]
+    solid_srp["SOLID-SRP<br/><small>Robert C. Martin</small>"]
+    clean_architecture["Clean Architecture<br/><small>Robert C. Martin</small>"]
+    vertical_slice_architecture["Vertical Slice Architecture<br/><small>Jimmy Bogard</small>"]
+    mikado_method["Mikado Method<br/><small>Ola Ellnestam</small>"]
+    spike_solution["Spike Solution<br/><small>Kent Beck</small>"]
+    thin_vertical_slice["Thin Vertical Slice<br/><small>Alistair Cockburn</small>"]
+    xy_problem["XY Problem Avoidance<br/><small>Mark Jason Dominus</small>"]
+    cynefin_framework["Cynefin Framework<br/><small>Dave Snowden</small>"]
+    wardley_mapping["Wardley Mapping<br/><small>Simon Wardley</small>"]
+    jobs_to_be_done["Jobs To Be Done<br/><small>Clayton Christensen</small>"]
+    occams_razor["Occam's Razor<br/><small>William of Ockham</small>"]
+    first_principles_thinking["First Principles Thinking<br/><small>Aristotle</small>"]
+    five_whys["Five Whys<br/><small>Taiichi Ohno</small>"]
+    feynman_technique["Feynman Technique<br/><small>Richard Feynman</small>"]
+    morphological_box["Morphological Box<br/><small>Fritz Zwicky</small>"]
+    swot["SWOT<br/><small>Albert Humphrey</small>"]
+    pugh_matrix["Pugh Matrix<br/><small>Stuart Pugh</small>"]
+    mece["MECE<br/><small>Barbara Minto</small>"]
+    ears_requirements["EARS<br/><small>Alistair Mavin</small>"]
+    invest["INVEST<br/><small>Bill Wake</small>"]
+    cockburn_use_cases["Cockburn Use Cases<br/><small>Alistair Cockburn</small>"]
+    prd["PRD<br/><small>Product Management Convention</small>"]
+    devils_advocate["Devil's Advocate<br/><small>Catholic Canonization Process</small>"]
+    goodharts_law["Goodhart's Law<br/><small>Charles Goodhart</small>"]
+    pert["PERT<br/><small>US Navy</small>"]
+    adr_according_to_nygard["ADR<br/><small>Michael Nygard</small>"]
+    quality_attribute_scenario["Quality Attribute Scenario<br/><small>Software Architecture Convention</small>"]
+    moscow["MoSCoW<br/><small>Dai Clegg</small>"]
+  end
+  subgraph prove["PROVE"]
+    chain_of_thought["Chain-of-Thought Reasoning<br/><small>Wei</small>"]
+  end
+  subgraph emit["EMIT"]
+    dry["DRY<br/><small>Andy Hunt</small>"]
+    kiss_principle["KISS Principle<br/><small>Kelly Johnson</small>"]
+    yagni["YAGNI<br/><small>Ron Jeffries</small>"]
+    single_level_of_abstraction_principle["SLAP<br/><small>Kent Beck</small>"]
+    law_of_demeter["Law of Demeter<br/><small>Ian Holland</small>"]
+    code_smells["Code Smells<br/><small>Kent Beck</small>"]
+    cohesion_criteria["Cohesion Criteria<br/><small>Larry Constantine</small>"]
+    iosp["IOSP<br/><small>Ralf Westphal</small>"]
+    mental_model_according_to_naur["Programming as Theory Building<br/><small>Peter Naur</small>"]
+    sota["SOTA<br/><small>General Convention</small>"]
+    effective_go["Effective Go<br/><small>The Go Team</small>"]
+    conways_law["Conway's Law<br/><small>Melvin Conway</small>"]
+    grasp["GRASP<br/><small>Craig Larman</small>"]
+    solid_dip["SOLID-DIP<br/><small>Robert C. Martin</small>"]
+    hexagonal_architecture["Hexagonal Architecture<br/><small>Alistair Cockburn</small>"]
+    arc42["arc42<br/><small>Peter Hruschka</small>"]
+    cap_theorem["CAP Theorem<br/><small>Eric Brewer</small>"]
+    fallacies_of_distributed_computing["Fallacies of Distributed Computing<br/><small>Peter Deutsch</small>"]
+    event_driven_architecture["Event-Driven Architecture<br/><small>Distributed Systems Convention</small>"]
+    walking_skeleton["Walking Skeleton<br/><small>Alistair Cockburn</small>"]
+    gof_facade_pattern["GoF-Facade<br/><small>Gamma Helm Johnson Vlissides</small>"]
+    gof_adapter_pattern["GoF-Adapter<br/><small>Gamma Helm Johnson Vlissides</small>"]
+    gof_chain_of_responsibility_pattern["GoF-Chain of Responsibility<br/><small>Gamma Helm Johnson Vlissides</small>"]
+    gof_observer_pattern["GoF-Observer<br/><small>Gamma Helm Johnson Vlissides</small>"]
+    gof_strategy_pattern["GoF-Strategy<br/><small>Gamma Helm Johnson Vlissides</small>"]
+    bem_methodology["BEM Methodology<br/><small>Yandex</small>"]
+    conventional_commits["Conventional Commits<br/><small>Community Specification</small>"]
+    github_flow["GitHub Flow<br/><small>GitHub</small>"]
+  end
+  subgraph state["STATE"]
+    fagan_inspection["Fagan Inspection<br/><small>Michael Fagan</small>"]
+    property_based_testing["Property-Based Testing<br/><small>Koen Claessen</small>"]
+    mutation_testing["Mutation Testing<br/><small>Richard Lipton</small>"]
+    red_green_tdd["Red/Green TDD<br/><small>Kent Beck</small>"]
+    tdd_chicago_school["TDD Chicago School<br/><small>Chicago/Detroit Tradition</small>"]
+    test_double_meszaros["Test Double<br/><small>Gerard Meszaros</small>"]
+    testing_pyramid["Testing Pyramid<br/><small>Mike Cohn</small>"]
+  end
+  subgraph conc["CONC"]
+  end
+  subgraph sec["SEC"]
+    owasp_top_10["OWASP Top 10<br/><small>OWASP Foundation</small>"]
+    stride["STRIDE Threat Model<br/><small>Loren Kohnfelder</small>"]
+    postels_law["Postel's Law<br/><small>Jon Postel</small>"]
+    linddun["LINDDUN Privacy Threat Model<br/><small>KU Leuven</small>"]
+    iec_61508_sil_levels["IEC 61508 SIL Levels<br/><small>IEC</small>"]
+    regulated_environment["Regulated Environment<br/><small>Compliance Convention</small>"]
+  end
+  subgraph res["RES"]
+    site_reliability_engineering["Site Reliability Engineering<br/><small>Ben Treynor</small>"]
+  end
+  subgraph decide["DECIDE"]
+    definition_of_done["Definition of Done<br/><small>Ken Schwaber</small>"]
+    llm_evaluations["LLM-Evaluations<br/><small>LLM Evaluation Practice</small>"]
+    iso_25010["ISO/IEC 25010<br/><small>ISO</small>"]
+    control_chart_shewhart["Control Chart<br/><small>Walter Shewhart</small>"]
+    nelson_rules["Nelson Rules<br/><small>Lloyd S. Nelson</small>"]
+    spc["SPC<br/><small>Walter Shewhart / W. Edwards Deming</small>"]
+  end
+  subgraph update_docs["UPDATE_DOCS"]
+    pyramid_principle["Pyramid Principle<br/><small>Barbara Minto</small>"]
+    bluf["BLUF<br/><small>US Military Doctrine</small>"]
+    inverted_pyramid_style["Inverted Pyramid Style<br/><small>Journalism Convention</small>"]
+    plain_english_strunk_white["Plain English<br/><small>William Strunk Jr</small>"]
+    aida_model["AIDA Model<br/><small>E. St. Elmo Lewis</small>"]
+    hemingway_bridge["Hemingway Bridge<br/><small>Ernest Hemingway</small>"]
+    diataxis_framework["Diataxis Framework<br/><small>Daniele Procida</small>"]
+    docs_as_code["Docs-as-Code<br/><small>Ralf D. Muller</small>"]
+    blooms_taxonomy["Bloom's Taxonomy<br/><small>Benjamin Bloom</small>"]
+  end
+  subgraph gaps["ADDED (missing from reference)"]
+    big_o_algorithmic_complexity["Big O Algorithmic Complexity<br/><small>Donald Knuth</small>"]:::gap
+    data_oriented_design["Data-Oriented Design<br/><small>Mike Acton</small>"]:::gap
+    mechanical_sympathy["Mechanical Sympathy<br/><small>Martin Thompson</small>"]:::gap
+    zero_cost_abstractions["Zero-Cost Abstractions<br/><small>Bjarne Stroustrup</small>"]:::gap
+    circuit_breaker["Circuit Breaker<br/><small>Michael Nygard</small>"]:::gap
+    mental_model_illegal_states["Make Illegal States Unrepresentable<br/><small>Yaron Minsky</small>"]:::gap
+  end
+  big_o_algorithmic_complexity -.->|belongs to| conc
+  data_oriented_design -.->|belongs to| conc
+  mechanical_sympathy -.->|belongs to| conc
+  zero_cost_abstractions -.->|belongs to| conc
+  circuit_breaker -.->|belongs to| res
+  mental_model_illegal_states -.->|belongs to| state
+
+  dry -.-> single_level_of_abstraction_principle
+  dry -.-> kiss_principle
+  dry -.-> yagni
+  kiss_principle -.-> yagni
+  kiss_principle -.-> solid_principles
+  yagni -.-> tdd_chicago_school
+  solid_principles -.-> clean_architecture
+  cynefin_framework -.-> wardley_mapping
+  xy_problem -.-> bluf
+  five_whys -.-> xy_problem
+  five_whys -.-> first_principles_thinking
+  mece -.-> pyramid_principle
+  mece -.-> bluf
+  mece -.-> morphological_box
+  adr_according_to_nygard -.-> arc42
+  mikado_method -.-> tdd_chicago_school
+  spike_solution -.-> walking_skeleton
+  spike_solution -.-> pugh_matrix
+  clean_architecture -.-> hexagonal_architecture
+  solid_srp -.-> solid_principles
+  solid_srp -.-> single_level_of_abstraction_principle
+  arc42 -.-> quality_attribute_scenario
+  conways_law -.-> cohesion_criteria
+  conways_law -.-> vertical_slice_architecture
+  grasp -.-> solid_principles
+  grasp -.-> clean_architecture
+  gof_observer_pattern -.-> gof_strategy_pattern
+  conventional_commits -.-> github_flow
+  conventional_commits -.-> definition_of_done
+  chain_of_thought -.-> first_principles_thinking
+  chain_of_thought -.-> feynman_technique
+  owasp_top_10 -.-> regulated_environment
+  owasp_top_10 -.-> iec_61508_sil_levels
+  stride -.-> owasp_top_10
+  stride -.-> regulated_environment
+  definition_of_done -.-> moscow
+  pyramid_principle -.-> bluf
+  pyramid_principle -.-> inverted_pyramid_style
+  bluf -.-> inverted_pyramid_style
+  bluf -.-> plain_english_strunk_white
+  occams_razor -.-> kiss_principle
+  occams_razor -.-> yagni
+  occams_razor -.-> five_whys
+  occams_razor -.-> mece
+  occams_razor -.-> devils_advocate
+  first_principles_thinking -.-> feynman_technique
+  wardley_mapping -.-> swot
+  invest -.-> moscow
+  ears_requirements -.-> cockburn_use_cases
+  ears_requirements -.-> invest
+  cockburn_use_cases -.-> arc42
+  cockburn_use_cases -.-> iso_25010
+  pert -.-> moscow
+  goodharts_law -.-> llm_evaluations
+  property_based_testing -.-> mutation_testing
+  property_based_testing -.-> testing_pyramid
+  mutation_testing -.-> testing_pyramid
+  red_green_tdd -.-> tdd_chicago_school
+  testing_pyramid -.-> tdd_chicago_school
+  fagan_inspection -.-> mutation_testing
+  fagan_inspection -.-> testing_pyramid
+  docs_as_code -.-> diataxis_framework
+  docs_as_code -.-> arc42
+  docs_as_code -.-> conventional_commits
+  diataxis_framework -.-> arc42
+  diataxis_framework -.-> inverted_pyramid_style
+  morphological_box -.-> pugh_matrix
+  swot -.-> pugh_matrix
+  swot -.-> moscow
+  devils_advocate -.-> five_whys
+  site_reliability_engineering -.-> five_whys
+  site_reliability_engineering -.-> spc
+  site_reliability_engineering -.-> control_chart_shewhart
+  vertical_slice_architecture -.-> clean_architecture
+  vertical_slice_architecture -.-> hexagonal_architecture
+  thin_vertical_slice -.-> walking_skeleton
+  thin_vertical_slice -.-> vertical_slice_architecture
+  walking_skeleton -.-> clean_architecture
+  walking_skeleton -.-> hexagonal_architecture
+  fallacies_of_distributed_computing -.-> cap_theorem
+  fallacies_of_distributed_computing -.-> event_driven_architecture
+  fallacies_of_distributed_computing -.-> hexagonal_architecture
+  cap_theorem -.-> event_driven_architecture
+  event_driven_architecture -.-> hexagonal_architecture
+  event_driven_architecture -.-> clean_architecture
+  postels_law -.-> solid_principles
+  postels_law -.-> event_driven_architecture
+  linddun -.-> owasp_top_10
+  linddun -.-> regulated_environment
+
+  classDef gap fill:#ffe4b3,stroke:#c77700,stroke-width:2px
+```
+
+## Reading the graph
+
+- **Solid subgraph membership** = the gm phase whose prose already names this anchor (`gm-config/prose/<phase>.md`, `Preferences (named, narrow)`).
+- **Dotted edges** = a real `:related:` backreference pulled live from `llm-coding/Semantic-Anchors`'s `docs/anchors/<id>.adoc` source, not invented. An edge existing means the two anchors compose: a proposal citing one should check the other.
+- **`gaps` subgraph (orange)** = well-known, well-attributed techniques in real literature that gm's prose already depends on (CONC's whole performance vocabulary; RES's Circuit Breaker; STATE's illegal-states principle) but that are absent from the public Semantic-Anchors catalog as of this graph's construction. `CONC`'s own subgraph is empty for exactly this reason -- every anchor CONC currently names is a gap, not yet catalogued upstream.
+
+## The 6 gaps, for upstream contribution
+
+| id | name | author | belongs to |
+| --- | --- | --- | --- |
+| `big-o-algorithmic-complexity` | Big O Algorithmic Complexity | Donald Knuth | CONC |
+| `data-oriented-design` | Data-Oriented Design | Mike Acton | CONC |
+| `mechanical-sympathy` | Mechanical Sympathy | Martin Thompson | CONC |
+| `zero-cost-abstractions` | Zero-Cost Abstractions | Bjarne Stroustrup | CONC |
+| `circuit-breaker` | Circuit Breaker | Michael Nygard | RES |
+| `mental-model-illegal-states` | Make Illegal States Unrepresentable | Yaron Minsky | STATE |
+
+These are candidates for a PR against `llm-coding/Semantic-Anchors` (`docs/anchors/<id>.adoc`, following the existing frontmatter shape: `:categories:`, `:roles:`, `:proponents:`, `:tags:`, `:related:`, `:tier:`, `:definition:`), not a fork -- the catalog is a shared reference, and gm benefits from every project's anchor set staying converged on one upstream source rather than diverging per project.
+
+## Using this graph to prime self-reconfiguration
+
+When `self-reconfig-candidate` fires (a gate denial repeating past `policy.gate_repeat_escalate_threshold`), before composing an `fsm-propose-override` proposal:
+
+1. Identify which phase the friction occurred in.
+2. Walk that phase's subgraph here for an anchor that already names the missing discipline.
+3. Follow its dotted edges one hop -- the adjacent anchors are the ones a reviewer will expect the proposal to also account for.
+4. If nothing in the graph names the gap, that is itself signal: the friction is either genuinely novel (write the override in gm's own voice) or names a technique missing from both gm's prose and the upstream catalog (add it to the gaps table above, and consider it for the same upstream PR).
+
+A proposal that cites an anchor id from this graph is reviewable against real, external, authored literature. A proposal that doesn't is asking a reviewer to trust prose alone -- exactly the gap `fsm-propose-override`'s `AskUserQuestion` authority guard exists to catch for graph/hook-bearing overrides, and the same discipline is worth applying informally to prose-only overrides too.