bmad-module-skill-forge 2.1.0 → 3.0.0

package/src/skf-test-skill/scripts/gap-ledger.py

@@ -0,0 +1,1381 @@
+#!/usr/bin/env python3
+# /// script
+# requires-python = ">=3.11"
+# dependencies = []
+# ///
+"""SKF test-skill gap ledger: one typed JSON record per finding.
+
+test-skill's hard gate (step-hard-gate.md) must see every Critical and High
+finding before scoring, and update-skill must read every gap with its source
+and remediation after a FAIL. Markdown written by the model serves neither:
+the coverage section holds tables and counts, the coherence formats carry no
+severity, and the Gap Entry Format was applied only in the report step, after
+the gate. This script keeps the findings in a JSON ledger instead. Each stage
+appends the gaps it finds, classified against the Gap Severity table in
+scoring-rules.md when they are written; hard-gate.py counts the blocking ones,
+and the Gap Report is rendered from the same records on every path.
+
+Ledger file: `{forge_version}/test-findings-{run_id}.json`
+
+  {
+    "schema_version": 1,
+    "run_id": "<run id from the file name>" | null,
+    "stages": ["coverage-check", "coherence-check", ...],
+    "records": [
+      {
+        "id": "GAP-001",               # assigned here, in append order
+        "severity": "Critical" | "High" | "Medium" | "Low" | "Info",
+        "category": "<slug from CATEGORIES>",
+        "group": "Coverage" | "Coherence" | "Structural" | "Discovery" | "External",
+        "title": "<one line>",
+        "source": "<file:line or a section reference>",
+        "remediation": "<exact action that fixes the gap>",
+        "issue": "<what is wrong>",    # optional
+        "export": "<export name>",     # optional
+        "stage": "<the stage that appended it>"
+      }
+    ]
+  }
+
+`stages` lists every stage that appended, even with no records, so the gate
+can tell a stage that found nothing from a stage that never ran. `group` is
+derived from the category; callers never supply `id`, `group` or `stage`.
+
+Subcommands:
+
+  append --ledger <path> --stage <stage> [--input <file>]...
+      Read records from each <file> in turn, or from stdin without --input:
+      a JSON array, one record object, or {"records": [...]}. An empty array
+      records the stage with no findings. Every record is validated first
+      and nothing is written unless all are valid. A record equal to one
+      already in the ledger (same severity, category, title, source and
+      export) is not added again, so a stage that runs twice leaves one
+      copy. Creates the ledger when it does not exist.
+      Appends may overlap (parallel tool calls): each holds an exclusive
+      lock on <ledger>.skf-lock from its read to its write, and another
+      append waits for it (up to 30 s, then LEDGER_LOCKED), so no record an
+      append reported is lost. The write is atomic: a temp file of its own
+      beside the ledger, then a rename. The lock file is removed afterwards.
+      Output: {"status": "ok", "ledger", "stage", "appended": [ids],
+               "duplicates": [ids], "record_count"}
+
+  append --ledger <path> --stage <stage> --from <kind> --input <file>...
+         [--surface <file>] [--signatures <file>] [--numerator <file>]
+         [--stale <file>] [--provenance <file>] [--skill-dir <dir>]
+         [--metadata <path>] [--served <family>]...
+      Build the records from the result file a script wrote (see Adapters
+      below) instead of reading them, then append them as above, with the
+      same output. A gap a script finds is recorded with a fixed title,
+      Source, `export` and remediation, so a rerun dedupes it. Each kind
+      reads one --input (structure reads one or more) and only the options
+      its entry names; any other option, or a file that does not hold the
+      result its kind reads, is INVALID_INPUT and nothing is written.
+
+Adapters (--from), each record classified by its Gap Severity row
+(scoring-rules.md). `{meta}` is --metadata, the metadata.json path a record
+cites (never read; default `metadata.json`):
+
+  coverage  --input reconcile-coverage.py's result; --surface, --signatures,
+            --numerator, --stale, --provenance, --skill-dir, --metadata.
+            Barrel branch: each `missing` name is a Medium `missing-export`
+            gap `Missing export: {name}`, or a Medium `missing-type` gap
+            `Missing type: {name}` when the --signatures result lists it in
+            `missingTypes`; its Source is the file and line the --surface
+            result's `exports[]` records for it, else `{meta}`. Each `stale`
+            name the --stale result (classify-stale) marks `fabricated` is a
+            Critical `fabricated-signature` gap `Fabricated signature:
+            {name}` at its `source` (one with no `source` is INVALID_INPUT,
+            never a lesser gap); any other is a Medium
+            `stale-documentation` gap `Stale documentation: {name}` at the
+            first line of SKILL.md, then of references/*.md in path order,
+            that writes the name (validate-inventory.py's match, with
+            --skill-dir), else `SKILL.md`. Scalar and stack branches: a
+            `missingCount` above 0 is one Medium `missing-export` gap
+            `{missingCount} of {denominator} exports not documented` at
+            `{meta}`, except that a scalar count taken from the verified
+            numerator (`numeratorSource` `verified`) is replaced by one
+            `Missing export: {name}` gap per --numerator `absent` name, at
+            the first --provenance entry of that name, else `{meta}`.
+            Every per-name gap carries `export`.
+  guards    --input load-coverage-inputs.py's surface result; --metadata.
+            `guards.deflation.fires` is a Medium `metadata-drift` gap at
+            `{meta}` (else the result's `inputs.metadata`), and
+            `guards.inflation.fires` a Medium `denominator-inflation` gap at
+            the result's `inputs.brief` (else `skill-brief.yaml`), whose
+            remediation recommends `stats.effective_denominator` when
+            `guards.umbrella.umbrella` is true, else `scope.tier_a_include`.
+  numerator --input verify-declared-numerator.py's result; --metadata.
+            `inflated` true is a High `numerator-inflation` gap at `{meta}`
+            whose issue lists the `absent` names.
+  metadata-coherence
+            --input check-metadata-coherence.py's result; --metadata. Each
+            `findings[]` entry, titled as it is and with its `detail` as the
+            issue, at `{meta}`: Medium is a Medium `metadata-drift` gap, Info
+            an Info `multi-denominator` gap.
+  provenance-line
+            --input skf-verify-provenance-completeness.py verify's result.
+            Each `stale[]` item whose reason is `line-not-definition`, at
+            its `{source_file}:{source_line}` with `export`: a Low
+            `provenance-line` gap `Provenance line is not the definition of
+            {export_name}` when `definition_lines` holds a line, else an Info
+            `provenance-unverified` gap `Provenance line not verified for
+            {export_name}`. Its other items are not gaps here.
+  coherence --input aggregate-coherence.py's result. Each
+            `invalidReferences[]` entry at `SKILL.md:{line}`, by `status`:
+            `missing` is a Critical `broken-reference` gap, `inaccurate` a
+            High `inaccurate-reference` gap (its `issues` as the issue) and
+            `escapes` a High `reference-escape` gap naming its `canonical`
+            path.
+  structure --input one or more skf-scan-skill-md-structure.py results,
+            each told apart by its keys; --served. A `scan --required-sections`
+            result: each family not `satisfied` is a High `structural` gap
+            `naive-coherence: missing required section: {family}`, unless
+            --served names it (another heading serves it: the caller's
+            judgment) or it is `description` and `frontmatter_description`
+            is true. A `scan` result: unbalanced fences are a High
+            `structural` gap, and each bare opening fence and each
+            `table_drift` row a Medium `structural` gap at `SKILL.md:{line}`.
+            A `usage-scope` result: each `zero_usage` export is a Medium
+            `structural` gap with `export`. A `scan` or `reference-check`
+            result whose `scripts_assets.missing` is true is a Medium
+            `scripts-assets` gap. The titles are coherence-check.md's
+            `naive-coherence:` texts; a `reference-check` result (contextual
+            mode) titles its gap `coherence:` instead.
+
+  render --ledger <path> [--heading]
+      Print the Gap Report body in Markdown on stdout: the totals, the
+      Remediation Summary table and one Gap Entry per record, Critical first,
+      then by id. Each entry names its category as `{group} ({category})`,
+      which skf-parse-gaps.py reads back for a report whose ledger is gone.
+      --heading adds the `## Gap Report` line. With no records the body says
+      no gaps were found.
+
+  summary --ledger <path>
+      Output: {"status": "ok", "ledger", "run_id", "stages", "total",
+               "counts": {severity: n}, "blocking", "non_blocking",
+               "by_category": {category: n}}
+
+A Critical or High record blocks only when a stage before the hard gate
+appended it. The report stage appends after the gate (the discovery test),
+so render and summary count its Critical and High records as non-blocking:
+render lists them on a line of their own, after the hard gate.
+
+  categories
+      Output the category vocabulary and the severities:
+      {"status": "ok", "severities": [...],
+       "categories": [{"category", "group", "description"}]}
+
+Exit codes:
+  0  the result was printed
+  1  the ledger is missing, unreadable, not a valid ledger, held by another
+     append past the wait, or not writable
+  2  the input records are invalid (nothing was written), or a usage error
+     (argparse: a missing or unknown argument, usage on stderr, no JSON)
+
+Errors print {"status": "error", "code", "error", ...} on stdout, except for
+render, which prints it on stderr so a redirected Gap Report stays clean. The
+codes: LEDGER_MISSING, LEDGER_INVALID, LEDGER_LOCKED, WRITE_FAILED (exit 1),
+INVALID_INPUT and INVALID_RECORD (exit 2, with `errors: [{index, error}]`).
+"""
+
+from __future__ import annotations
+
+import argparse
+import contextlib
+import errno
+import importlib.util
+import json
+import os
+import re
+import secrets
+import sys
+import time
+from pathlib import Path
+
+if os.name == "nt":
+    import msvcrt
+else:
+    import fcntl
+
+SCHEMA_VERSION = 1
+
+# How long an append waits for another append's lock before LEDGER_LOCKED.
+# An append holds it for milliseconds; the wait covers a burst of parallel
+# tool calls, not a writer that hangs.
+LOCK_TIMEOUT_SECONDS = 30.0
+_LOCK_POLL_SECONDS = 0.02
+# Windows refuses to replace a file another process holds open, and a virus
+# scanner or the search indexer may open a just-written ledger for a moment:
+# save_ledger retries the rename there for up to this long before failing.
+_IS_WINDOWS = os.name == "nt"
+_REPLACE_WAIT_SECONDS = 5.0
+
+SEVERITIES = ("Critical", "High", "Medium", "Low", "Info")
+BLOCKING = frozenset({"Critical", "High"})
+_SEVERITY_BY_LOWER = {s.lower(): s for s in SEVERITIES}
+# Stages that append after the hard gate decided (report.md records the
+# discovery test there): a Critical or High gap they record blocked nothing.
+POST_GATE_STAGES = frozenset({"report"})
+
+# Category slug -> (group, what the category covers). update-skill routes a
+# gap on its category, not its severity, so the slugs are a closed set: a typo
+# fails the append instead of reaching update-skill as an unknown category.
+CATEGORIES: dict[str, tuple[str, str]] = {
+    "missing-export": ("Coverage", "an exported function or class the skill does not document"),
+    "signature-mismatch": ("Coverage", "a documented signature that differs from the source"),
+    "fabricated-signature": ("Coverage", "a documented export that is absent at the cited source line"),
+    "stale-documentation": ("Coverage", "documentation for an export the source no longer has"),
+    "missing-type": ("Coverage", "a type or interface the skill does not document"),
+    "provenance-completeness": ("Coverage", "an export the skill documents that the provenance map lacks"),
+    "provenance-line": ("Coverage", "a provenance line that is not the definition of its export"),
+    "provenance-unverified": ("Coverage", "a provenance line the line-check rules could not verify"),
+    "metadata-drift": ("Coverage", "export counts in the metadata that diverge from each other"),
+    "denominator-inflation": ("Coverage", "a scope include union larger than the provenance map"),
+    "numerator-inflation": (
+        "Coverage",
+        "a documented count equal to the denominator while declared exports are absent from the skill",
+    ),
+    "multi-denominator": ("Coverage", "barrel and documented-surface counts that diverge by design"),
+    "migration-section": ("Coverage", "a migration section that disagrees with the annotation data"),
+    "broken-reference": ("Coherence", "a reference whose target does not exist"),
+    "inaccurate-reference": ("Coherence", "a reference whose target exists but does not match it"),
+    "reference-escape": (
+        "Coherence",
+        "a reference whose real path leaves the skill, its source and, for a stack, the skills folder",
+    ),
+    "integration-pattern": ("Coherence", "an integration pattern that is incomplete"),
+    "split-body-mismatch": ("Coherence", "SKILL.md and a references/ file that disagree about an export"),
+    "scripts-assets": ("Coherence", "a Scripts & Assets section that is missing or names a missing file"),
+    "scripts-assets-provenance": ("Coherence", "a script or asset file without a provenance entry"),
+    "structural": (
+        "Structural",
+        "a section, code fence, table or usage example in SKILL.md that is missing or wrong",
+    ),
+    "metadata": ("Structural", "missing optional metadata or examples"),
+    "observation": ("Structural", "a style suggestion or another non-blocking observation"),
+    "discovery": ("Discovery", "a discovery test that misrouted prompts or did not run"),
+    "description": ("Discovery", "a description whose triggers need optimizing"),
+    "external-validator": ("External", "a skill-check diagnostic or a Tessl Review suggestion"),
+}
+
+REQUIRED_FIELDS = ("severity", "category", "title", "source", "remediation")
+OPTIONAL_FIELDS = ("issue", "export")
+ASSIGNED_FIELDS = ("id", "group", "stage")
+
+# The Remediation Summary's Estimated Effort column, per severity.
+EFFORT = {
+    "Critical": "Read the source code and write or correct the documentation",
+    "High": "Read the source code and write or correct the documentation",
+    "Medium": "Add type definitions, interface docs or the missing section",
+    "Low": "Add examples or metadata",
+    "Info": "Optional improvements, no action required",
+}
+
+_ID_RE = re.compile(r"^GAP-(\d{3,})$")
+_STAGE_RE = re.compile(r"^[a-z][a-z0-9-]*$")
+_RUN_ID_FROM_NAME = re.compile(r"^test-findings-(.+)\.json$")
+
+
+class LedgerError(Exception):
+    """A ledger that is missing, unreadable or malformed."""
+
+    def __init__(self, code: str, message: str) -> None:
+        super().__init__(message)
+        self.code = code
+
+
+# --------------------------------------------------------------------------
+# Records
+# --------------------------------------------------------------------------
+
+
+def _one_line(value: str) -> str:
+    return " ".join(value.split())
+
+
+def normalize_record(raw: object) -> tuple[dict | None, str | None]:
+    """Validate one input record. Returns (record, None) or (None, error).
+
+    >>> rec, err = normalize_record({"severity": "high", "category": "broken-reference",
+    ...     "title": " Broken  link ", "source": "SKILL.md:12", "remediation": "Fix it."})
+    >>> (rec["severity"], rec["title"], err)
+    ('High', 'Broken link', None)
+    >>> normalize_record({"severity": "Severe"})[1]
+    "missing required field 'category'"
+    """
+    if not isinstance(raw, dict):
+        return None, f"a record must be a JSON object, got {type(raw).__name__}"
+    for key in ASSIGNED_FIELDS:
+        if key in raw:
+            return None, f"'{key}' is assigned by the ledger; leave it out"
+    unknown = sorted(set(raw) - set(REQUIRED_FIELDS) - set(OPTIONAL_FIELDS))
+    if unknown:
+        return None, f"unknown field(s): {', '.join(unknown)}"
+    for key in REQUIRED_FIELDS:
+        if key not in raw:
+            return None, f"missing required field '{key}'"
+        if not isinstance(raw[key], str):
+            return None, f"'{key}' must be a string"
+    for key in OPTIONAL_FIELDS:
+        if raw.get(key) is not None and not isinstance(raw[key], str):
+            return None, f"'{key}' must be a string or null"
+    severity = _SEVERITY_BY_LOWER.get(raw["severity"].strip().lower())
+    if severity is None:
+        return None, (
+            f"unknown severity '{raw['severity']}' (expected one of {', '.join(SEVERITIES)})"
+        )
+    category = raw["category"].strip()
+    if category not in CATEGORIES:
+        return None, f"unknown category '{category}' (run the categories subcommand for the list)"
+    record = {
+        "severity": severity,
+        "category": category,
+        "group": CATEGORIES[category][0],
+        "title": _one_line(raw["title"]),
+        "source": _one_line(raw["source"]),
+        "remediation": raw["remediation"].strip(),
+    }
+    for key in ("title", "source", "remediation"):
+        if not record[key]:
+            return None, f"'{key}' must not be empty"
+    issue = (raw.get("issue") or "").strip()
+    if issue:
+        record["issue"] = issue
+    export = _one_line(raw.get("export") or "")
+    if export:
+        record["export"] = export
+    return record, None
+
+
+def _dedupe_key(record: dict) -> tuple[str, str, str, str, str | None]:
+    """What makes two records the same gap. The export is part of it: two
+    missing exports can share a generic title and a barrel Source."""
+    return (
+        record["severity"],
+        record["category"],
+        record["title"],
+        record["source"],
+        record.get("export") or None,
+    )
+
+
+def _id_number(record: dict) -> int:
+    m = _ID_RE.match(record.get("id", ""))
+    return int(m.group(1)) if m else 0
+
+
+def ordered(records: list[dict]) -> list[dict]:
+    """Records by severity (Critical first), then by id."""
+    rank = {s: i for i, s in enumerate(SEVERITIES)}
+    return sorted(records, key=lambda r: (rank[r["severity"]], _id_number(r)))
+
+
+def count_by_severity(records: list[dict]) -> dict[str, int]:
+    counts = {s: 0 for s in SEVERITIES}
+    for record in records:
+        counts[record["severity"]] += 1
+    return counts
+
+
+def is_blocking(record: dict) -> bool:
+    """A Critical or High gap that a stage before the hard gate recorded.
+
+    >>> is_blocking({"severity": "High", "stage": "coherence-check"})
+    True
+    >>> is_blocking({"severity": "High", "stage": "report"})
+    False
+    """
+    return record["severity"] in BLOCKING and record.get("stage") not in POST_GATE_STAGES
+
+
+# --------------------------------------------------------------------------
+# Ledger file
+# --------------------------------------------------------------------------
+
+
+def new_ledger(path: Path) -> dict:
+    m = _RUN_ID_FROM_NAME.match(path.name)
+    return {
+        "schema_version": SCHEMA_VERSION,
+        "run_id": m.group(1) if m else None,
+        "stages": [],
+        "records": [],
+    }
+
+
+def validate_ledger(data: object) -> None:
+    """Raise LedgerError unless `data` is a well-formed ledger."""
+    if not isinstance(data, dict):
+        raise LedgerError("LEDGER_INVALID", "the ledger must be a JSON object")
+    if data.get("schema_version") != SCHEMA_VERSION:
+        raise LedgerError(
+            "LEDGER_INVALID",
+            f"unsupported schema_version {data.get('schema_version')!r} (expected {SCHEMA_VERSION})",
+        )
+    stages = data.get("stages")
+    if not isinstance(stages, list) or not all(isinstance(s, str) for s in stages):
+        raise LedgerError("LEDGER_INVALID", "'stages' must be a list of strings")
+    records = data.get("records")
+    if not isinstance(records, list):
+        raise LedgerError("LEDGER_INVALID", "'records' must be a list")
+    seen: set[str] = set()
+    for index, record in enumerate(records):
+        if not isinstance(record, dict):
+            raise LedgerError("LEDGER_INVALID", f"record {index} is not an object")
+        rid = record.get("id")
+        if not isinstance(rid, str) or not _ID_RE.match(rid):
+            raise LedgerError("LEDGER_INVALID", f"record {index} has no valid id")
+        if rid in seen:
+            raise LedgerError("LEDGER_INVALID", f"duplicate id {rid}")
+        seen.add(rid)
+        if record.get("severity") not in SEVERITIES:
+            raise LedgerError("LEDGER_INVALID", f"{rid} has an unknown severity")
+        if record.get("category") not in CATEGORIES:
+            raise LedgerError("LEDGER_INVALID", f"{rid} has an unknown category")
+        for key in ("title", "source", "remediation"):
+            if not isinstance(record.get(key), str) or not record[key]:
+                raise LedgerError("LEDGER_INVALID", f"{rid} has no {key}")
+        for key in OPTIONAL_FIELDS:
+            if record.get(key) is not None and not isinstance(record[key], str):
+                raise LedgerError("LEDGER_INVALID", f"{rid} has a {key} that is not a string")
+
+
+def load_ledger(path: Path) -> dict:
+    """Read and validate a ledger. Raises LedgerError."""
+    if not path.is_file():
+        raise LedgerError("LEDGER_MISSING", f"no ledger at {path}")
+    try:
+        data = json.loads(path.read_text(encoding="utf-8"))
+    except (OSError, UnicodeDecodeError) as exc:
+        raise LedgerError("LEDGER_INVALID", f"cannot read {path}: {exc}") from exc
+    except json.JSONDecodeError as exc:
+        raise LedgerError("LEDGER_INVALID", f"{path} is not valid JSON: {exc}") from exc
+    validate_ledger(data)
+    return data
+
+
+def save_ledger(path: Path, data: dict) -> None:
+    """Write the ledger atomically: a temp file of its own beside it, then a
+    rename. Call it under ledger_lock() so no other append runs between the
+    read and this write."""
+    path.parent.mkdir(parents=True, exist_ok=True)
+    tmp = path.with_name(f"{path.name}.{os.getpid()}-{secrets.token_hex(4)}.skf-tmp")
+    payload = (json.dumps(data, indent=2, ensure_ascii=False) + "\n").encode("utf-8")
+    # O_EXCL: never write into another writer's temp file. O_BINARY (Windows
+    # only; 0 elsewhere) keeps the \n line ends.
+    flags = os.O_WRONLY | os.O_CREAT | os.O_EXCL | getattr(os, "O_BINARY", 0)
+    fd = os.open(tmp, flags, 0o644)
+    try:
+        with open(fd, "wb") as fh:
+            fh.write(payload)
+            fh.flush()
+            os.fsync(fh.fileno())
+        deadline = time.monotonic() + _REPLACE_WAIT_SECONDS
+        while True:
+            try:
+                os.replace(tmp, path)
+                break
+            except PermissionError:
+                if not _IS_WINDOWS or time.monotonic() >= deadline:
+                    raise
+                time.sleep(_LOCK_POLL_SECONDS)
+    except BaseException:
+        with contextlib.suppress(OSError):
+            tmp.unlink()
+        raise
+
+
+def _try_lock(fd: int) -> bool:
+    """Lock fd exclusively without waiting; False while another process holds it."""
+    try:
+        if os.name == "nt":
+            msvcrt.locking(fd, msvcrt.LK_NBLCK, 1)
+        else:
+            fcntl.flock(fd, fcntl.LOCK_EX | fcntl.LOCK_NB)
+    except OSError as exc:
+        if exc.errno in (errno.EAGAIN, errno.EACCES, errno.EDEADLK):
+            return False
+        raise
+    return True
+
+
+def _unlock(fd: int) -> None:
+    with contextlib.suppress(OSError):
+        if os.name == "nt":
+            msvcrt.locking(fd, msvcrt.LK_UNLCK, 1)
+        else:
+            fcntl.flock(fd, fcntl.LOCK_UN)
+
+
+def _names_open_file(fd: int, path: Path) -> bool:
+    """True when `path` still names the file open as `fd`."""
+    try:
+        return os.path.samestat(os.fstat(fd), os.stat(path))
+    except OSError:
+        return False
+
+
+@contextlib.contextmanager
+def ledger_lock(path: Path, timeout: float | None = None):
+    """Hold the exclusive append lock on `<ledger>.skf-lock`.
+
+    The same fcntl / msvcrt lock as skf-atomic-write.py flip-link, but this
+    one waits: up to `timeout` seconds (LOCK_TIMEOUT_SECONDS), then it raises
+    LedgerError LEDGER_LOCKED. The lock dies with its process, so a crashed
+    append leaves nothing to clean up. The lock file is removed on release.
+    On POSIX that happens while the lock is still held, and a waiter checks
+    that the file it locked is still the one at the path: otherwise a waiter
+    on the removed file and a newcomer on a new one would both get a lock.
+    Windows removes no file that is open, so there the file goes after the
+    release, unless a waiter has it open.
+    """
+    timeout = LOCK_TIMEOUT_SECONDS if timeout is None else timeout
+    lock_path = path.with_name(path.name + ".skf-lock")
+    path.parent.mkdir(parents=True, exist_ok=True)
+    deadline = time.monotonic() + timeout
+
+    def refuse() -> LedgerError:
+        return LedgerError("LEDGER_LOCKED", f"another append held {lock_path} for more than {timeout:g} s")
+
+    while True:
+        fd = os.open(lock_path, os.O_WRONLY | os.O_CREAT, 0o644)
+        try:
+            while not _try_lock(fd):
+                if time.monotonic() >= deadline:
+                    raise refuse()
+                time.sleep(_LOCK_POLL_SECONDS)
+            if os.name == "nt" or _names_open_file(fd, lock_path):
+                break
+            _unlock(fd)
+        except BaseException:
+            os.close(fd)
+            raise
+        os.close(fd)
+        if time.monotonic() >= deadline:
+            raise refuse()
+    try:
+        yield
+    finally:
+        if os.name != "nt":
+            with contextlib.suppress(OSError):
+                lock_path.unlink()
+        _unlock(fd)
+        os.close(fd)
+        if os.name == "nt":
+            with contextlib.suppress(OSError):
+                lock_path.unlink()
+
+
+def append_records(ledger: dict, stage: str, records: list[dict]) -> tuple[list[str], list[str]]:
+    """Append validated records in place. Returns (appended ids, duplicate ids)."""
+    existing = {_dedupe_key(r): r["id"] for r in ledger["records"]}
+    next_number = max((_id_number(r) for r in ledger["records"]), default=0) + 1
+    appended: list[str] = []
+    duplicates: list[str] = []
+    for record in records:
+        key = _dedupe_key(record)
+        if key in existing:
+            duplicates.append(existing[key])
+            continue
+        rid = f"GAP-{next_number:03d}"
+        next_number += 1
+        ledger["records"].append({"id": rid, **record, "stage": stage})
+        existing[key] = rid
+        appended.append(rid)
+    if stage not in ledger["stages"]:
+        ledger["stages"].append(stage)
+    return appended, duplicates
+
+
+def parse_input(text: str) -> list:
+    """The records an append reads: an array, one object or {"records": [...]}."""
+    data = json.loads(text)
+    if isinstance(data, list):
+        return data
+    if isinstance(data, dict):
+        if set(data) == {"records"}:
+            if not isinstance(data["records"], list):
+                raise ValueError("'records' must be a list")
+            return data["records"]
+        return [data]
+    raise ValueError("input must be a JSON array or object")
+
+
+# --------------------------------------------------------------------------
+# Adapters: a script's result file -> records (append --from)
+# --------------------------------------------------------------------------
+
+
+class AdapterError(Exception):
+    """A --from input that does not hold the result its kind reads."""
+
+
+FAMILIES = ("description", "usage", "api_surface")
+# The options (argparse dests) each kind reads, besides --input. A kind
+# refuses any other, so a call that passes a file it would ignore fails.
+ADAPTER_OPTIONS: dict[str, frozenset[str]] = {
+    "coverage": frozenset(
+        {"surface", "signatures", "numerator", "stale", "provenance", "skill_dir", "metadata"}
+    ),
+    "guards": frozenset({"metadata"}),
+    "numerator": frozenset({"metadata"}),
+    "metadata-coherence": frozenset({"metadata"}),
+    "provenance-line": frozenset(),
+    "coherence": frozenset(),
+    "structure": frozenset({"served"}),
+}
+ALL_ADAPTER_OPTIONS = frozenset().union(*ADAPTER_OPTIONS.values())
+_DEFAULT_METADATA = "metadata.json"
+_DEFAULT_BRIEF = "skill-brief.yaml"
+# The remediation of a missing required section: the SKF template heading
+# that serves the family.
+_SECTION_FIX = {
+    "description": (
+        "Add a `## Overview` section to SKILL.md that says what the library does and when "
+        "to use it, or a non-empty `description` to its frontmatter."
+    ),
+    "usage": (
+        "Add a `## Quick Start` section to SKILL.md with a runnable example of the "
+        "library's most common calls."
+    ),
+    "api_surface": (
+        "Add a `## Key API Summary` section to SKILL.md that lists the library's exports "
+        "with their purpose."
+    ),
+}
+
+_SIBLINGS: dict[str, object] = {}
+
+
+def _sibling(filename: str):
+    """A script beside this one, loaded once: validate-inventory.py (the
+    documented-name match the stale-documentation Source uses)."""
+    module = _SIBLINGS.get(filename)
+    if module is None:
+        path = Path(__file__).resolve().parent / filename
+        spec = importlib.util.spec_from_file_location("skf_" + filename[:-3].replace("-", "_"), path)
+        if spec is None or spec.loader is None or not path.is_file():
+            raise AdapterError(f"{filename} not found beside {Path(__file__).name}")
+        module = importlib.util.module_from_spec(spec)
+        spec.loader.exec_module(module)
+        _SIBLINGS[filename] = module
+    return module
+
+
+def read_result(path: str, flag: str):
+    """The JSON a script wrote. Raises AdapterError."""
+    try:
+        return json.loads(Path(path).read_bytes().decode("utf-8-sig"))
+    except (OSError, UnicodeDecodeError) as exc:
+        raise AdapterError(f"cannot read {flag} {path}: {exc}") from exc
+    except json.JSONDecodeError as exc:
+        raise AdapterError(f"{flag} {path} is not JSON: {exc.msg}") from exc
+
+
+def _object(data: object, label: str, *keys: str) -> dict:
+    """`data` as the object a script writes, with each of `keys`."""
+    if not isinstance(data, dict) or any(key not in data for key in keys):
+        needs = ", ".join(f"'{key}'" for key in keys)
+        raise AdapterError(f"{label} is not that script's result: it needs an object with {needs}")
+    return data
+
+
+def _list(data: dict, key: str, label: str) -> list:
+    value = data.get(key)
+    if value is None:
+        return []
+    if not isinstance(value, list):
+        raise AdapterError(f"{label}: '{key}' must be a list")
+    return value
+
+
+def _names(data: dict, key: str, label: str) -> list[str]:
+    return [n for n in _list(data, key, label) if isinstance(n, str) and n]
+
+
+def _count(value: object) -> int | None:
+    return value if isinstance(value, int) and not isinstance(value, bool) else None
+
+
+def _code(names: list[str]) -> str:
+    return ", ".join(f"`{n}`" for n in names)
+
+
+def _gap(severity: str, category: str, title: str, source: str, remediation: str,
+         issue: str | None = None, export: str | None = None) -> dict:
+    record = {"severity": severity, "category": category, "title": title, "source": source,
+              "remediation": remediation}
+    if issue:
+        record["issue"] = issue
+    if export:
+        record["export"] = export
+    return record
+
+
+def _surface_sources(surface: dict | None) -> dict[str, str]:
+    """name -> `file:line` (or `file`) from a surface result's `exports[]`:
+    the first record with a file, one with a line preferred."""
+    sources: dict[str, tuple[bool, str]] = {}
+    for rec in (surface or {}).get("exports") or []:
+        if not isinstance(rec, dict) or not isinstance(rec.get("name"), str):
+            continue
+        file = rec.get("file")
+        if not isinstance(file, str) or not file:
+            continue
+        line = _count(rec.get("line"))
+        where = (line is not None, f"{file}:{line}" if line is not None else file)
+        known = sources.get(rec["name"])
+        if known is None or (where[0] and not known[0]):
+            sources[rec["name"]] = where
+    return {name: where for name, (_, where) in sources.items()}
+
+
+def _provenance_sources(provenance: dict | None) -> dict[str, str]:
+    """export_name -> `source_file:source_line` of its first provenance entry."""
+    sources: dict[str, str] = {}
+    for entry in (provenance or {}).get("entries") or []:
+        if not isinstance(entry, dict) or not isinstance(entry.get("export_name"), str):
+            continue
+        file = entry.get("source_file")
+        if entry["export_name"] in sources or not isinstance(file, str) or not file:
+            continue
+        line = _count(entry.get("source_line"))
+        sources[entry["export_name"]] = f"{file}:{line}" if line is not None else file
+    return sources
+
+
+def _documenting_line(skill_dir: Path, name: str) -> str | None:
+    """The first line of SKILL.md, then of references/*.md in path order,
+    that writes `name` (validate-inventory.py's match), as `file:line`."""
+    pattern = _sibling("validate-inventory.py").name_pattern(name)
+    refs = skill_dir / "references"
+    files = [skill_dir / "SKILL.md"] + (sorted(refs.glob("*.md")) if refs.is_dir() else [])
+    for path in files:
+        try:
+            text = path.read_bytes().decode("utf-8-sig", errors="replace")
+        except OSError:
+            continue
+        for number, line in enumerate(text.split("\n"), start=1):
+            if pattern.search(line):
+                return f"{path.relative_to(skill_dir).as_posix()}:{number}"
+    return None
+
+
+def _missing_export(name: str, where: str | None, meta: str, kind: str = "export") -> dict:
+    if where:
+        read = f"read its definition at `{where}`"
+    else:
+        read = f"find its definition in the source (`{meta}` lists it)"
+    if kind == "type":
+        return _gap("Medium", "missing-type", f"Missing type: {name}", where or meta,
+                    f"Document the type `{name}` in SKILL.md: {read} and list its fields or members.",
+                    issue=f"the type `{name}` is on the source API surface, and the skill does not document it",
+                    export=name)
+    return _gap("Medium", "missing-export", f"Missing export: {name}", where or meta,
+                f"Document `{name}` in SKILL.md: {read} and add its signature, its purpose and a usage "
+                "example.",
+                issue=f"`{name}` is on the source API surface, and SKILL.md and references/ do not document it",
+                export=name)
+
+
+def from_coverage(cov: object, surface: object = None, signatures: object = None, numerator: object = None,
+                  stale: object = None, provenance: object = None, skill_dir: str | None = None,
+                  metadata: str | None = None) -> list[dict]:
+    """reconcile-coverage.py's result: missing and stale names, or the missing count."""
+    cov = _object(cov, "--input (reconcile-coverage.py)", "branch")
+    meta = metadata or _DEFAULT_METADATA
+    branch = cov["branch"]
+    records: list[dict] = []
+    if branch == "enumerated":
+        types = set()
+        if signatures is not None:
+            types = set(_names(_object(signatures, "--signatures (score-signatures.py)", "missingTypes"),
+                               "missingTypes", "--signatures"))
+        where = _surface_sources(_object(surface, "--surface (load-coverage-inputs.py surface)", "exports")
+                                 if surface is not None else None)
+        for name in _names(cov, "missing", "--input"):
+            records.append(_missing_export(name, where.get(name), meta, "type" if name in types else "export"))
+        classified: dict[str, dict] = {}
+        if stale is not None:
+            if not isinstance(stale, list):
+                raise AdapterError("--stale is not classify-stale JSON (a list)")
+            classified = {i["name"]: i for i in stale if isinstance(i, dict) and isinstance(i.get("name"), str)}
+        for name in _names(cov, "stale", "--input"):
+            item = classified.get(name) or {}
+            cited = item.get("source")
+            if item.get("fabricated") is True:
+                if not isinstance(cited, str) or not cited:
+                    # a Critical gap needs the line it cites: never downgrade it
+                    raise AdapterError(f"--stale: {name!r} is fabricated with no source")
+                records.append(_gap(
+                    "Critical", "fabricated-signature", f"Fabricated signature: {name}", cited,
+                    f"Remove `{name}` from SKILL.md and references/, or correct it to the export the source "
+                    f"defines: no file its provenance cites defines it (`{cited}`).",
+                    issue=f"SKILL.md documents `{name}`, which the source API surface lacks and the cited "
+                          "file does not define",
+                    export=name))
+                continue
+            doc = _documenting_line(Path(skill_dir), name) if skill_dir else None
+            records.append(_gap(
+                "Medium", "stale-documentation", f"Stale documentation: {name}", doc or "SKILL.md",
+                f"Check `{name}` against the source and update or remove its documentation at "
+                f"{doc or 'SKILL.md'}: the enumerated source surface does not export it.",
+                issue=f"SKILL.md documents `{name}`, which the enumerated source API surface lacks",
+                export=name))
+    elif branch in ("scalar", "stack"):
+        if branch == "scalar" and cov.get("numeratorSource") == "verified":
+            if numerator is None:
+                raise AdapterError("the scalar count used the verified numerator: pass its result with "
+                                   "--numerator")
+            absent = _names(_object(numerator, "--numerator (verify-declared-numerator.py)", "absent"),
+                            "absent", "--numerator")
+            where = _provenance_sources(_object(provenance, "--provenance (provenance map)", "entries")
+                                        if provenance is not None else None)
+            records += [_missing_export(name, where.get(name), meta) for name in absent]
+        else:
+            missing, denominator = _count(cov.get("missingCount")), _count(cov.get("denominator"))
+            if missing:
+                records.append(_gap(
+                    "Medium", "missing-export", f"{missing} of {denominator} exports not documented", meta,
+                    f"Find the {missing} exports the denominator counts that SKILL.md and references/ do not "
+                    f"name (`{meta}` lists the exports) and document each with its signature.",
+                    issue=f"SKILL.md and references/ name {cov.get('documented')} of the {denominator} exports "
+                          f"the {branch} denominator counts"))
+    elif branch != "docsOnly":
+        raise AdapterError(f"--input (reconcile-coverage.py): unknown branch {branch!r}")
+    return records
+
+
+def from_guards(surface: object, metadata: str | None = None) -> list[dict]:
+    """load-coverage-inputs.py surface's deflation and inflation guards."""
+    surface = _object(surface, "--input (load-coverage-inputs.py surface)", "guards", "inputs")
+    guards = surface["guards"]
+    if guards is None:
+        return []
+    if not isinstance(guards, dict):
+        raise AdapterError("--input (load-coverage-inputs.py surface): 'guards' must be an object or null")
+    inputs = surface["inputs"] if isinstance(surface["inputs"], dict) else {}
+    meta = metadata or inputs.get("metadata") or _DEFAULT_METADATA
+    brief = inputs.get("brief") or _DEFAULT_BRIEF
+    records: list[dict] = []
+    deflation = guards.get("deflation") or {}
+    if deflation.get("fires") is True:
+        records.append(_gap(
+            "Medium", "metadata-drift",
+            "denominator deflation: effective_denominator below source public surface without tier_a_include",
+            meta,
+            f"Set `stats.effective_denominator` in `{meta}` to the public surface the source re-derives, or "
+            "add `scope.tier_a_include` to the brief to name the authored tier the smaller count measures.",
+            issue=f"the re-derived surface counts {deflation.get('rederived')} exports, "
+                  f"{deflation.get('pct')}% above `stats.effective_denominator` "
+                  f"({deflation.get('effectiveDenominator')}), and the brief has no `scope.tier_a_include`"))
+    inflation = guards.get("inflation") or {}
+    if inflation.get("fires") is True:
+        if (guards.get("umbrella") or {}).get("umbrella") is True:
+            fix = (f"Set `stats.effective_denominator` in `{meta}` to the authored surface: the root barrel "
+                   "is an umbrella of re-exports, so a `scope.tier_a_include` glob would still count them.")
+        else:
+            fix = (f"Add `scope.tier_a_include` to `{brief}`, listing the files of the authored surface, so "
+                   "the denominator counts it instead of the coarse `scope.include` union.")
+        records.append(_gap(
+            "Medium", "denominator-inflation",
+            "denominator inflation: coarse scope.include union exceeds authored surface", brief, fix,
+            issue=f"the `scope.include` union counts {inflation.get('scopeIncludeUnion')} exports, "
+                  f"{inflation.get('pct')}% above the {inflation.get('provenanceEntries')} provenance entries, "
+                  "and the brief has no `scope.tier_a_include`"))
+    return records
+
+
+def from_numerator(numerator: object, metadata: str | None = None) -> list[dict]:
+    """verify-declared-numerator.py's result: the numerator inflation gap."""
+    numerator = _object(numerator, "--input (verify-declared-numerator.py)", "inflated")
+    if numerator["inflated"] is not True:
+        return []
+    meta = metadata or _DEFAULT_METADATA
+    declared, verified = _count(numerator.get("declared")) or 0, _count(numerator.get("verified")) or 0
+    absent = _names(numerator, "absent", "--input")
+    return [_gap(
+        "High", "numerator-inflation",
+        f"numerator inflation: {declared - verified} of {declared} declared exports absent from "
+        "SKILL.md/references",
+        meta,
+        f"Document the absent exports in SKILL.md, or set `stats.exports_documented` in `{meta}` to the "
+        f"{verified} exports the skill documents.",
+        issue=f"`stats.exports_documented` equals the denominator, and these declared exports are absent "
+              f"from SKILL.md and references/: {_code(absent)}")]
+
+
+def from_metadata_coherence(result: object, metadata: str | None = None) -> list[dict]:
+    """check-metadata-coherence.py's findings: count drift and multi-denominator notes."""
+    result = _object(result, "--input (check-metadata-coherence.py)", "findings")
+    meta = metadata or _DEFAULT_METADATA
+    records: list[dict] = []
+    for index, finding in enumerate(_list(result, "findings", "--input")):
+        if not isinstance(finding, dict) or not isinstance(finding.get("title"), str):
+            raise AdapterError(f"--input: findings[{index}] has no title")
+        detail = finding.get("detail") if isinstance(finding.get("detail"), str) else None
+        if finding.get("severity") == "Medium":
+            records.append(_gap(
+                "Medium", "metadata-drift", finding["title"], meta,
+                f"Recount the exports and correct the diverging counts in `{meta}`; update-skill applies it as "
+                "a metadata update.",
+                issue=detail))
+        elif finding.get("severity") == "Info":
+            records.append(_gap(
+                "Info", "multi-denominator", finding["title"], meta,
+                "No action required: the barrel and the documented surface measure different sets by design, "
+                "and the report records both counts.",
+                issue=detail))
+        else:
+            raise AdapterError(f"--input: findings[{index}] has severity {finding.get('severity')!r}, "
+                               "neither Medium nor Info")
+    return records
+
+
+def from_provenance_line(result: object) -> list[dict]:
+    """verify's `line-not-definition` items: provenance line gaps."""
+    result = _object(result, "--input (skf-verify-provenance-completeness.py verify)", "stale")
+    records: list[dict] = []
+    for item in _list(result, "stale", "--input"):
+        if not isinstance(item, dict) or item.get("reason") != "line-not-definition":
+            continue
+        name, file, line = item.get("export_name"), item.get("source_file"), item.get("source_line")
+        defs = [n for n in item.get("definition_lines") or [] if _count(n) is not None]
+        where = f"{file}:{line}"
+        if defs:
+            lines = ", ".join(str(n) for n in defs)
+            records.append(_gap(
+                "Low", "provenance-line", f"Provenance line is not the definition of {name}", where,
+                f"Set the provenance `source_line` of `{name}` in `{file}` to its definition line ({lines}) "
+                "and move its citations to that line; update-skill `--from-test-report` applies this when the "
+                "file defines it on one line.",
+                issue=f"the provenance map records `{name}` at `{where}`, which is not the line that defines "
+                      f"it; the lines that define it are {lines}.",
+                export=name))
+        else:
+            records.append(_gap(
+                "Info", "provenance-unverified", f"Provenance line not verified for {name}", where,
+                f"Open `{file}` and confirm that line {line} defines `{name}`; if another line does, set the "
+                "provenance `source_line` to it.",
+                issue=f"the line-check rules found no definition line for `{name}` in `{file}`; check by hand",
+                export=name))
+    return records
+
+
+def from_coherence(result: object) -> list[dict]:
+    """aggregate-coherence.py's invalid references."""
+    result = _object(result, "--input (aggregate-coherence.py)", "invalidReferences")
+    records: list[dict] = []
+    for index, ref in enumerate(_list(result, "invalidReferences", "--input")):
+        if not isinstance(ref, dict):
+            raise AdapterError(f"--input: invalidReferences[{index}] is not an object")
+        target, line = ref.get("target"), _count(ref.get("line"))
+        where = f"SKILL.md:{line}" if line is not None else "SKILL.md"
+        status = ref.get("status")
+        if status == "missing":
+            records.append(_gap(
+                "Critical", "broken-reference", f"coherence: broken reference: {target}", where,
+                f"Point the reference to `{target}` at {where} at a target that exists, or remove it: "
+                "nothing exists at its target.",
+                issue="; ".join(i for i in ref.get("issues") or [] if isinstance(i, str)) or None))
+        elif status == "inaccurate":
+            issues = "; ".join(i for i in ref.get("issues") or [] if isinstance(i, str))
+            records.append(_gap(
+                "High", "inaccurate-reference", f"coherence: inaccurate reference: {target}", where,
+                f"Correct the reference to `{target}` at {where} so it matches its target"
+                + (f": {issues}." if issues else "."),
+                issue=issues or None))
+        elif status == "escapes":
+            canonical = ref.get("canonical")
+            records.append(_gap(
+                "High", "reference-escape",
+                f"coherence: reference escapes skill/source sandbox: {target} → {canonical}", where,
+                f"Point the reference to `{target}` at {where} inside the skill package or its source tree, "
+                f"or remove it: it resolves to `{canonical}`."))
+        else:
+            raise AdapterError(f"--input: invalidReferences[{index}] has an unknown status {status!r}")
+    return records
+
+
+def _structure_shape(data: object) -> str | None:
+    """Which skf-scan-skill-md-structure.py result `data` is."""
+    if not isinstance(data, dict):
+        return None
+    if all(isinstance(data.get(f), dict) and "satisfied" in data[f] for f in FAMILIES):
+        return "required-sections"
+    if "zero_usage" in data:
+        return "usage-scope"
+    if "unbalanced_fences" in data:
+        return "scan"
+    if "references" in data and "counts" in data:
+        return "reference-check"
+    return None
+
+
+def _scripts_assets_gap(data: dict, prefix: str) -> list[dict]:
+    section = data.get("scripts_assets")
+    if not isinstance(section, dict) or section.get("missing") is not True:
+        return []
+    folders = [f for f in section.get("folders") or [] if isinstance(f, str)]
+    named = " and ".join(f"`{f}/`" for f in folders) or "`scripts/` and `assets/`"
+    return [_gap(
+        "Medium", "scripts-assets",
+        f"{prefix}: scripts/assets directory exists but Scripts & Assets section missing", "SKILL.md",
+        f"Add a `## Scripts & Assets` section to SKILL.md that lists each file in {named} with its purpose.")]
+
+
+def from_structure(results: list[object], served: list[str] | None = None) -> list[dict]:
+    """skf-scan-skill-md-structure.py results: the naive structural gaps, and
+    the Scripts & Assets gap of either mode."""
+    served_set = set(served or [])
+    records: list[dict] = []
+    for index, data in enumerate(results):
+        shape = _structure_shape(data)
+        if shape is None:
+            raise AdapterError(f"--input #{index + 1} is not a scan, scan --required-sections, usage-scope or "
+                               "reference-check result")
+        if shape == "required-sections":
+            for family in FAMILIES:
+                entry = data[family]
+                if entry.get("satisfied") is True or family in served_set:
+                    continue
+                if family == "description" and data.get("frontmatter_description") is True:
+                    continue
+                tried = [t for t in entry.get("tried") or [] if isinstance(t, str)]
+                records.append(_gap(
+                    "High", "structural", f"naive-coherence: missing required section: {family}", "SKILL.md",
+                    _SECTION_FIX[family],
+                    issue=f"no heading of SKILL.md serves the {family} family (synonyms tried: "
+                          f"{', '.join(tried)})"))
+        elif shape == "scan":
+            if data.get("unbalanced_fences") is True:
+                records.append(_gap(
+                    "High", "structural", "naive-coherence: unbalanced code fence (unclosed block)", "SKILL.md",
+                    "Close the code block SKILL.md leaves open with a fence line of three backticks.",
+                    issue=f"SKILL.md has {data.get('fence_count')} fence lines, so one block is never closed"))
+            for fence in _list(data, "bare_opening_fences", "--input"):
+                line = _count(fence.get("line")) if isinstance(fence, dict) else None
+                records.append(_gap(
+                    "Medium", "structural",
+                    f"naive-coherence: opening code fence at line {line} missing language tag",
+                    f"SKILL.md:{line}",
+                    f"Add a language tag (such as `bash` or `python`) to the opening fence at SKILL.md line "
+                    f"{line}."))
+            for row in _list(data, "table_drift", "--input"):
+                if not isinstance(row, dict):
+                    continue
+                line, expected = _count(row.get("line")), _count(row.get("expected_cols"))
+                where = f"section `{row['section']}`: " if row.get("section") else ""
+                records.append(_gap(
+                    "Medium", "structural",
+                    f"naive-coherence: table row at line {line} has {row.get('actual_cols')} columns; "
+                    f"header has {expected}",
+                    f"SKILL.md:{line}",
+                    f"Give the row at SKILL.md line {line} the {expected} columns its header has (write a pipe "
+                    "inside a cell as `\\|`).",
+                    issue=f"{where}{row.get('row')}" if row.get("row") else None))
+            records += _scripts_assets_gap(data, "naive-coherence")
+        elif shape == "usage-scope":
+            for export in _list(data, "zero_usage", "--input"):
+                if not isinstance(export, dict) or not isinstance(export.get("name"), str):
+                    continue
+                name, kind = export["name"], export.get("kind") or "export"
+                records.append(_gap(
+                    "Medium", "structural",
+                    f"naive-coherence: exported {kind} `{name}` is not referenced in any usage-family "
+                    "section or reference file",
+                    "SKILL.md",
+                    f"Add an example that calls `{name}` to SKILL.md's usage section (or to a references/ file "
+                    "for a split body).",
+                    export=name))
+        else:
+            records += _scripts_assets_gap(data, "coherence")
+    return records
+
+
+def build_records(args: argparse.Namespace) -> list[dict]:
+    """The records `append --from` appends. Raises AdapterError."""
+    kind = args.adapter
+    stray = sorted(o for o in ALL_ADAPTER_OPTIONS - ADAPTER_OPTIONS[kind] if getattr(args, o) is not None)
+    if stray:
+        flags = ", ".join("--" + o.replace("_", "-") for o in stray)
+        raise AdapterError(f"--from {kind} does not read {flags}")
+    inputs = args.input or []
+    if not inputs or (kind != "structure" and len(inputs) > 1):
+        wanted = "one or more --input files" if kind == "structure" else "one --input file"
+        raise AdapterError(f"--from {kind} reads {wanted}")
+    if kind == "structure":
+        return from_structure([read_result(p, "--input") for p in inputs], args.served)
+    data = read_result(inputs[0], "--input")
+    if kind == "coverage":
+        files = {o: read_result(getattr(args, o), "--" + o) if getattr(args, o) is not None else None
+                 for o in ("surface", "signatures", "numerator", "stale", "provenance")}
+        return from_coverage(data, skill_dir=args.skill_dir, metadata=args.metadata, **files)
+    if kind == "guards":
+        return from_guards(data, args.metadata)
+    if kind == "numerator":
+        return from_numerator(data, args.metadata)
+    if kind == "metadata-coherence":
+        return from_metadata_coherence(data, args.metadata)
+    if kind == "provenance-line":
+        return from_provenance_line(data)
+    return from_coherence(data)
+
+
+# --------------------------------------------------------------------------
+# Gap Report
+# --------------------------------------------------------------------------
+
+
+def render_gap_report(ledger: dict, heading: bool = False) -> str:
+    """The Gap Report section body, from the ledger (report.md §4c writes it)."""
+    records = ordered(ledger["records"])
+    counts = count_by_severity(records)
+    blocking = sum(1 for r in records if is_blocking(r))
+    after_gate = counts["Critical"] + counts["High"] - blocking
+    lines: list[str] = []
+    if heading:
+        lines += ["## Gap Report", ""]
+    lines += [
+        f"**Total Gaps:** {len(records)}",
+        f"**Blocking (Critical + High):** {blocking}",
+        f"**Non-blocking (Medium + Low + Info):** {len(records) - blocking - after_gate}",
+    ]
+    if after_gate:
+        lines.append(f"**Found after the hard gate (Critical + High, non-blocking):** {after_gate}")
+    lines.append("")
+    if not records:
+        lines += ["No gaps found.", ""]
+        return "\n".join(lines)
+    lines += [
+        "### Remediation Summary",
+        "",
+        "| Severity | Count | Estimated Effort |",
+        "|----------|-------|------------------|",
+    ]
+    for severity in SEVERITIES:
+        effort = EFFORT[severity] if counts[severity] else "None"
+        lines.append(f"| {severity} | {counts[severity]} | {effort} |")
+    lines += [f"| **Total** | **{len(records)}** | |", ""]
+    for record in records:
+        lines += [
+            f"### {record['id']}: {record['title']}",
+            "",
+            f"**Severity:** {record['severity']}",
+            # The group comes from the category, as at append time: a ledger
+            # written by hand may lack it.
+            f"**Category:** {CATEGORIES[record['category']][0]} ({record['category']})",
+            f"**Source:** {record['source']}",
+        ]
+        if record.get("export"):
+            lines.append(f"**Export:** {record['export']}")
+        lines.append("")
+        if record.get("issue"):
+            lines += [f"**Issue:** {record['issue']}", ""]
+        lines += [f"**Remediation:** {record['remediation']}", ""]
+    return "\n".join(lines)
+
+
+def summarize(path: Path, ledger: dict) -> dict:
+    records = ledger["records"]
+    counts = count_by_severity(records)
+    by_category: dict[str, int] = {}
+    for record in records:
+        by_category[record["category"]] = by_category.get(record["category"], 0) + 1
+    blocking = sum(1 for r in records if is_blocking(r))
+    return {
+        "status": "ok",
+        "ledger": str(path),
+        "run_id": ledger.get("run_id"),
+        "stages": list(ledger["stages"]),
+        "total": len(records),
+        "counts": counts,
+        "blocking": blocking,
+        "non_blocking": len(records) - blocking,
+        "by_category": dict(sorted(by_category.items())),
+    }
+
+
+# --------------------------------------------------------------------------
+# CLI
+# --------------------------------------------------------------------------
+
+
+def _emit(payload: dict, stream=None) -> None:
+    print(json.dumps(payload, indent=2, ensure_ascii=False), file=stream or sys.stdout)
+
+
+def _error(code: str, message: str, **extra) -> dict:
+    return {"status": "error", "code": code, "error": message, **extra}
+
+
+def _read_records(args: argparse.Namespace) -> list:
+    """The records an append reads: built by --from, else from each --input
+    file in turn, else from stdin. Raises AdapterError, OSError or ValueError."""
+    if args.adapter:
+        return build_records(args)
+    given = sorted(o for o in ALL_ADAPTER_OPTIONS if getattr(args, o) is not None)
+    if given:
+        flags = ", ".join("--" + o.replace("_", "-") for o in given)
+        raise AdapterError(f"{flags} work only with --from")
+    if not args.input:
+        # A byte order mark on stdin (a Windows editor or shell) is not JSON.
+        return parse_input(sys.stdin.read().removeprefix("\ufeff"))
+    records: list = []
+    for path in args.input:
+        records += parse_input(Path(path).read_text(encoding="utf-8-sig"))
+    return records
+
+
+def _cmd_append(args: argparse.Namespace) -> int:
+    path = Path(args.ledger)
+    if not _STAGE_RE.match(args.stage):
+        _emit(_error("INVALID_INPUT", f"--stage must be a lowercase slug, got '{args.stage}'"))
+        return 2
+    try:
+        raw_records = _read_records(args)
+    except AdapterError as exc:
+        _emit(_error("INVALID_INPUT", str(exc)))
+        return 2
+    except (OSError, UnicodeDecodeError, ValueError) as exc:
+        # json.JSONDecodeError is a ValueError.
+        _emit(_error("INVALID_INPUT", f"cannot read the records: {exc}"))
+        return 2
+    records: list[dict] = []
+    errors: list[dict] = []
+    for index, raw in enumerate(raw_records):
+        record, err = normalize_record(raw)
+        if err:
+            errors.append({"index": index, "error": err})
+        else:
+            records.append(record)
+    if errors:
+        _emit(_error("INVALID_RECORD", f"{len(errors)} invalid record(s); nothing was written", errors=errors))
+        return 2
+    try:
+        # Read, append and write as one step: an append that overlaps this
+        # one waits here instead of writing over the records it adds.
+        with ledger_lock(path):
+            ledger = load_ledger(path) if path.exists() else new_ledger(path)
+            appended, duplicates = append_records(ledger, args.stage, records)
+            save_ledger(path, ledger)
+    except LedgerError as exc:
+        _emit(_error(exc.code, str(exc)))
+        return 1
+    except OSError as exc:
+        _emit(_error("WRITE_FAILED", f"cannot write {path}: {exc}"))
+        return 1
+    _emit(
+        {
+            "status": "ok",
+            "ledger": str(path),
+            "stage": args.stage,
+            "appended": appended,
+            "duplicates": duplicates,
+            "record_count": len(ledger["records"]),
+        }
+    )
+    return 0
+
+
+def _cmd_render(args: argparse.Namespace) -> int:
+    path = Path(args.ledger)
+    try:
+        ledger = load_ledger(path)
+    except LedgerError as exc:
+        _emit(_error(exc.code, str(exc)), stream=sys.stderr)
+        return 1
+    sys.stdout.write(render_gap_report(ledger, heading=args.heading))
+    return 0
+
+
+def _cmd_summary(args: argparse.Namespace) -> int:
+    path = Path(args.ledger)
+    try:
+        ledger = load_ledger(path)
+    except LedgerError as exc:
+        _emit(_error(exc.code, str(exc)))
+        return 1
+    _emit(summarize(path, ledger))
+    return 0
+
+
+def _cmd_categories(args: argparse.Namespace) -> int:
+    _emit(
+        {
+            "status": "ok",
+            "severities": list(SEVERITIES),
+            "categories": [
+                {"category": slug, "group": group, "description": description}
+                for slug, (group, description) in CATEGORIES.items()
+            ],
+        }
+    )
+    return 0
+
+
+def _build_parser() -> argparse.ArgumentParser:
+    parser = argparse.ArgumentParser(
+        prog="gap-ledger",
+        description="Append, render and summarize the test-skill gap ledger.",
+    )
+    sub = parser.add_subparsers(dest="cmd", required=True)
+
+    p = sub.add_parser("append", help="append records from stdin, --input or a script's result (--from)")
+    p.add_argument("--ledger", required=True, help="{forge_version}/test-findings-{run_id}.json")
+    p.add_argument("--stage", required=True, help="the stage appending, e.g. coverage-check")
+    p.add_argument("--input", action="append",
+                   help="a JSON file with the records (default: stdin); with --from, the script result to read "
+                        "(structure takes it more than once)")
+    p.add_argument("--from", dest="adapter", choices=tuple(ADAPTER_OPTIONS),
+                   help="build the records from the --input result of a script (see the docstring's Adapters)")
+    p.add_argument("--surface", help="coverage: load-coverage-inputs.py surface's result (Sources, kinds)")
+    p.add_argument("--signatures", help="coverage: score-signatures.py score's result (missingTypes)")
+    p.add_argument("--numerator", help="coverage: verify-declared-numerator.py's result (absent names)")
+    p.add_argument("--stale", help="coverage: classify-stale's result (fabricated signatures)")
+    p.add_argument("--provenance", help="coverage: the provenance map (an absent name's Source)")
+    p.add_argument("--skill-dir", help="coverage: the skill package (a stale name's documenting line)")
+    p.add_argument("--metadata", help="the metadata.json path a record cites as its Source (not read)")
+    p.add_argument("--served", action="append", choices=FAMILIES,
+                   help="structure: a family another heading serves (the caller's judgment); repeatable")
+    p.set_defaults(func=_cmd_append)
+
+    p = sub.add_parser("render", help="print the Gap Report in Markdown")
+    p.add_argument("--ledger", required=True)
+    p.add_argument("--heading", action="store_true", help="start with the ## Gap Report line")
+    p.set_defaults(func=_cmd_render)
+
+    p = sub.add_parser("summary", help="counts by severity and category")
+    p.add_argument("--ledger", required=True)
+    p.set_defaults(func=_cmd_summary)
+
+    p = sub.add_parser("categories", help="list the category vocabulary")
+    p.set_defaults(func=_cmd_categories)
+
+    return parser
+
+
+def _force_utf8(*streams) -> None:
+    """Reconfigure the JSON and Markdown streams to UTF-8.
+
+    A default Windows console uses cp1252, which cannot carry a non-ASCII
+    title or remediation. Keeps each stream's error handler; skips test
+    doubles without reconfigure(). For stdin this must run before the read.
+    """
+    for stream in streams:
+        if hasattr(stream, "reconfigure"):
+            errors = getattr(stream, "errors", None)
+            if errors is None:
+                stream.reconfigure(encoding="utf-8")
+            else:
+                stream.reconfigure(encoding="utf-8", errors=errors)
+
+
+def main(argv: list[str] | None = None) -> int:
+    _force_utf8(sys.stdin, sys.stdout, sys.stderr)
+    args = _build_parser().parse_args(argv)
+    return args.func(args)
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())