CST-agent/src/tools/web.c

#include <ctype.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <strings.h>

#include <curl/curl.h>

#include "cst_agent.h"
#include "cst_attach.h"
#include "cst_tools.h"
#include "cst_util.h"
#include "cst_web_md.h"
#include "os_compat.h"
#include "tools_internal.h"

#define FETCH_DOWNLOAD_MAX (5 * 1024 * 1024)
#define FETCH_DEFAULT_OUT (64 * 1024)
#define FETCH_MAX_OUT (1024 * 1024)

typedef struct {
  cst_buf_t body;
  int overflow;
} fetch_t;

static size_t on_body(char *p, size_t sz, size_t n, void *ud) {
  fetch_t *f = ud;
  size_t len = sz * n;
  if (f->body.len + len > FETCH_DOWNLOAD_MAX) {
    f->overflow = 1;
    return 0; /* abort */
  }
  cst_buf_append(&f->body, p, len);
  return len;
}

static int tag_is(const char *p, const char *name) {
  size_t n = strlen(name);
  if (strncasecmp(p, name, n) != 0)
    return 0;
  char c = p[n];
  return c == '>' || c == ' ' || c == '/' || c == '\t' || c == '\n' ||
         c == '\r';
}

static const char *const BLOCK_TAGS[] = {
    "p", "br", "div", "li", "tr", "h1", "h2", "h3", "h4", "h5", "h6",
    "section", "article", "header", "footer", "pre", "table", "ul", "ol",
    "blockquote", "hr"};

static void put_entity(cst_buf_t *b, const char **pp, const char *end) {
  static const struct {
    const char *name;
    const char *rep;
  } ents[] = {{"&amp;", "&"},  {"&lt;", "<"},    {"&gt;", ">"},
              {"&quot;", "\""}, {"&#39;", "'"},  {"&apos;", "'"},
              {"&nbsp;", " "}};
  const char *p = *pp;
  for (size_t i = 0; i < sizeof(ents) / sizeof(ents[0]); i++) {
    size_t n = strlen(ents[i].name);
    if ((size_t)(end - p) >= n && strncmp(p, ents[i].name, n) == 0) {
      cst_buf_puts(b, ents[i].rep);
      *pp = p + n;
      return;
    }
  }
  if (end - p > 3 && p[1] == '#') {
    char *e = NULL;
    long cp = p[2] == 'x' || p[2] == 'X' ? strtol(p + 3, &e, 16)
                                         : strtol(p + 2, &e, 10);
    if (e && *e == ';' && cp > 0 && cp < 0x110000) {
      char u[4];
      int k = 0;
      if (cp < 0x80) {
        u[k++] = (char)cp;
      } else if (cp < 0x800) {
        u[k++] = (char)(0xC0 | (cp >> 6));
        u[k++] = (char)(0x80 | (cp & 0x3F));
      } else if (cp < 0x10000) {
        u[k++] = (char)(0xE0 | (cp >> 12));
        u[k++] = (char)(0x80 | ((cp >> 6) & 0x3F));
        u[k++] = (char)(0x80 | (cp & 0x3F));
      } else {
        u[k++] = (char)(0xF0 | (cp >> 18));
        u[k++] = (char)(0x80 | ((cp >> 12) & 0x3F));
        u[k++] = (char)(0x80 | ((cp >> 6) & 0x3F));
        u[k++] = (char)(0x80 | (cp & 0x3F));
      }
      cst_buf_append(b, u, (size_t)k);
      *pp = e + 1;
      return;
    }
  }
  cst_buf_append(b, p, 1);
  *pp = p + 1;
}

/* Crude HTML → text: drop script/style/head, tags become spaces or
 * newlines, entities decoded, blank runs collapsed. */
static char *html_to_text(const char *s, size_t n) {
  cst_buf_t b = {0};
  const char *p = s, *end = s + n;
  for (const char *t = s; t + 7 < end; t++) {
    if (*t == '<' && strncasecmp(t + 1, "title", 5) == 0) {
      const char *gt = memchr(t, '>', (size_t)(end - t));
      const char *e = gt ? gt + 1 : end;
      while (e + 7 < end && strncasecmp(e, "</title", 7) != 0)
        e++;
      if (gt && e + 7 < end && e > gt + 1) {
        cst_buf_puts(&b, "# ");
        cst_buf_append(&b, gt + 1, (size_t)(e - gt - 1));
        cst_buf_puts(&b, "\n\n");
      }
      break;
    }
  }
  while (p < end) {
    if (*p == '<') {
      const char *q = p + 1;
      int closing = *q == '/';
      if (closing)
        q++;
      if (!closing && (tag_is(q, "script") || tag_is(q, "style") ||
                       tag_is(q, "head") || tag_is(q, "noscript"))) {
        const char *name = tag_is(q, "script")  ? "</script"
                           : tag_is(q, "style") ? "</style"
                           : tag_is(q, "head")  ? "</head"
                                                : "</noscript";
        const char *e = p;
        size_t nl = strlen(name);
        while (e < end && !(end - e >= (long)nl && strncasecmp(e, name, nl) == 0))
          e++;
        const char *gt = e < end ? memchr(e, '>', (size_t)(end - e)) : NULL;
        p = gt ? gt + 1 : end;
        continue;
      }
      if (end - p >= 4 && strncmp(p, "<!--", 4) == 0) {
        const char *e = p + 4;
        while (e + 2 < end && strncmp(e, "-->", 3) != 0)
          e++;
        p = e + 3 <= end ? e + 3 : end;
        continue;
      }
      int block = 0;
      for (size_t i = 0; i < sizeof(BLOCK_TAGS) / sizeof(BLOCK_TAGS[0]); i++)
        if (tag_is(q, BLOCK_TAGS[i]))
          block = 1;
      const char *gt = memchr(p, '>', (size_t)(end - p));
      p = gt ? gt + 1 : end;
      cst_buf_puts(&b, block ? "\n" : " ");
      continue;
    }
    if (*p == '&') {
      put_entity(&b, &p, end);
      continue;
    }
    cst_buf_append(&b, p, 1);
    p++;
  }
  /* collapse whitespace: spaces within lines, at most one blank line */
  cst_buf_t o = {0};
  int nls = 0, sp = 0;
  for (size_t i = 0; i < b.len; i++) {
    char ch = b.buf[i];
    if (ch == '\r')
      continue;
    if (ch == '\n') {
      if (nls < 2 && o.len)
        cst_buf_puts(&o, "\n");
      nls++;
      sp = 0;
      continue;
    }
    if (ch == ' ' || ch == '\t') {
      sp = 1;
      continue;
    }
    if (sp && nls == 0 && o.len)
      cst_buf_puts(&o, " ");
    sp = 0;
    nls = 0;
    cst_buf_append(&o, &ch, 1);
  }
  cst_buf_free(&b);
  if (!o.buf)
    cst_buf_puts(&o, "");
  return cst_buf_steal(&o);
}

static int textual(const char *ctype) {
  if (!ctype || !ctype[0])
    return 1;
  return strncasecmp(ctype, "text/", 5) == 0 || strstr(ctype, "json") ||
         strstr(ctype, "xml") || strstr(ctype, "javascript") ||
         strstr(ctype, "yaml") || strstr(ctype, "toml");
}

int cst_tool_web_fetch(const cst_tool_ctx_t *ctx, const ctl_tool_call_t *c,
                       cst_tool_result_t *r) {
  (void)ctx;
  const char *url = tool_arg(c, "url");
  if (!url || !url[0])
    return tool_set_result(r, "error: missing url", 1);
  while (*url == ' ')
    url++;
  if (strncasecmp(url, "http://", 7) != 0 &&
      strncasecmp(url, "https://", 8) != 0)
    return tool_set_result(r, "error: only http:// and https:// URLs", 1);
  long max_out = tool_arg_long(c, "max_bytes", FETCH_DEFAULT_OUT);
  if (max_out <= 0)
    max_out = FETCH_DEFAULT_OUT;
  if (max_out > FETCH_MAX_OUT)
    max_out = FETCH_MAX_OUT;

  CURL *curl = curl_easy_init();
  if (!curl)
    return tool_set_result(r, "error: curl init failed", 1);
  fetch_t f;
  memset(&f, 0, sizeof(f));
  char errbuf[CURL_ERROR_SIZE];
  errbuf[0] = 0;
  cst_curl_apply_ca(curl);
  curl_easy_setopt(curl, CURLOPT_URL, url);
  curl_easy_setopt(curl, CURLOPT_FOLLOWLOCATION, 1L);
  curl_easy_setopt(curl, CURLOPT_MAXREDIRS, 5L);
#if LIBCURL_VERSION_NUM >= 0x075500
  curl_easy_setopt(curl, CURLOPT_PROTOCOLS_STR, "http,https");
  curl_easy_setopt(curl, CURLOPT_REDIR_PROTOCOLS_STR, "http,https");
#else
  curl_easy_setopt(curl, CURLOPT_PROTOCOLS, (long)(CURLPROTO_HTTP | CURLPROTO_HTTPS));
  curl_easy_setopt(curl, CURLOPT_REDIR_PROTOCOLS,
                   (long)(CURLPROTO_HTTP | CURLPROTO_HTTPS));
#endif
  curl_easy_setopt(curl, CURLOPT_CONNECTTIMEOUT, 10L);
  curl_easy_setopt(curl, CURLOPT_TIMEOUT, 30L);
  curl_easy_setopt(curl, CURLOPT_NOSIGNAL, 1L);
  curl_easy_setopt(curl, CURLOPT_USERAGENT, "cst-agent/1.0 (+web_fetch)");
  curl_easy_setopt(curl, CURLOPT_ACCEPT_ENCODING, "");
  curl_easy_setopt(curl, CURLOPT_WRITEFUNCTION, on_body);
  curl_easy_setopt(curl, CURLOPT_WRITEDATA, &f);
  curl_easy_setopt(curl, CURLOPT_ERRORBUFFER, errbuf);
  CURLcode rc = curl_easy_perform(curl);
  long http = 0;
  char *ctype = NULL;
  char *final_url = NULL;
  curl_easy_getinfo(curl, CURLINFO_RESPONSE_CODE, &http);
  curl_easy_getinfo(curl, CURLINFO_CONTENT_TYPE, &ctype);
  curl_easy_getinfo(curl, CURLINFO_EFFECTIVE_URL, &final_url);
  cst_buf_t b = {0};
  int code = 0;
  if (rc != CURLE_OK && !f.overflow) {
    cst_buf_printf(&b, "error: fetch failed: %s",
                   errbuf[0] ? errbuf : curl_easy_strerror(rc));
    code = 1;
  } else {
    cst_buf_printf(&b, "HTTP %ld %s\ncontent-type: %s\n\n", http,
                   final_url ? final_url : url, ctype ? ctype : "?");
    if (http >= 400)
      code = 1;
    if (!textual(ctype)) {
      cst_buf_printf(&b, "(non-text content, %zu bytes not shown)\n",
                     f.body.len);
    } else {
      char *text = NULL;
      size_t tlen = 0;
      int html = ctype && strstr(ctype, "html");
      if (html) {
        text = html_to_text(f.body.buf ? f.body.buf : "", f.body.len);
        tlen = text ? strlen(text) : 0;
      }
      const char *src = html ? text : (f.body.buf ? f.body.buf : "");
      size_t slen = html ? tlen : f.body.len;
      size_t keep = slen > (size_t)max_out
                        ? cst_utf8_floor(src, (size_t)max_out)
                        : slen;
      size_t clean_len = 0;
      char *clean = cst_utf8_sanitize(src, keep, &clean_len);
      if (clean)
        cst_buf_append(&b, clean, clean_len);
      free(clean);
      if (keep < slen || f.overflow)
        cst_buf_printf(&b, "\n...(truncated; %zu%s bytes in total)\n", slen,
                       f.overflow ? "+" : "");
      free(text);
    }
  }
  curl_easy_cleanup(curl);
  cst_buf_free(&f.body);
  return tool_take_buf(r, &b, code);
}

/* --- md_to_html ---------------------------------------------------------- */

#define MD_TO_HTML_MAX (8 * 1024 * 1024)

/* The file name a URL ends in; "" when it names nothing (a bare host, a
 * directory). Kept to plain characters: it becomes a file name. */
static void name_from_url(const char *url, char *out, size_t n) {
  const char *p = strstr(url, "://");
  p = p ? p + 3 : url;
  const char *q = strpbrk(p, "?#");
  const char *end = q ? q : p + strlen(p);
  const char *slash = NULL;
  for (const char *s = p; s < end; s++)
    if (*s == '/')
      slash = s;
  const char *leaf = slash ? slash + 1 : p;
  size_t len = (size_t)(end - leaf);
  for (size_t i = 0; i < len; i++) {
    unsigned char ch = (unsigned char)leaf[i];
    if (!(isalnum(ch) || ch == '.' || ch == '-' || ch == '_' || ch == '+')) {
      len = i; /* cut at the first byte that is not a file name's */
      break;
    }
  }
  while (len && leaf[len - 1] == '.')
    len--;
  if (!len)
    snprintf(out, n, "download.bin");
  else
    snprintf(out, n, "%.*s", (int)len, leaf);
}

/* "docs/notes.md" -> "notes" (the title a document falls back to). */
static void stem_of(const char *rel, char *out, size_t n) {
  char leaf[128];
  snprintf(leaf, sizeof(leaf), "%.127s", rel && rel[0] ? rel : "");
  char *slash = NULL;
  for (char *p = leaf; *p; p++)
    if (*p == '/' || *p == '\\')
      slash = p;
  char *base = slash ? slash + 1 : leaf;
  char *dot = strrchr(base, '.');
  if (dot && dot != base)
    *dot = 0;
  snprintf(out, n, "%s", base[0] ? base : "document");
}

/* A title turned into a file name: what a path cannot carry becomes '-',
 * control bytes go, and a name that is only dots or spaces is refused. */
static void stem_of_title(const char *title, char *out, size_t n) {
  size_t i = 0;
  for (const char *p = title; *p && i + 1 < n && i < 80; p++) {
    unsigned char ch = (unsigned char)*p;
    if (ch == '/' || ch == '\\' || ch == ':' || ch == '*' || ch == '?' ||
        ch == '"' || ch == '<' || ch == '>' || ch == '|')
      ch = '-';
    else if (ch < 0x20)
      continue;
    out[i++] = (char)ch;
  }
  while (i && (out[i - 1] == ' ' || out[i - 1] == '.'))
    i--;
  size_t start = 0;
  while (start < i && (out[start] == ' ' || out[start] == '.'))
    start++;
  i -= start;
  if (i)
    memmove(out, out + start, i);
  out[i] = 0;
  if (!i)
    snprintf(out, n, "document");
}

/* The file md_to_html writes: `out` when given (a directory takes the
 * source's name inside it), otherwise the source file's stem, or the title. */
static void md_out_name(const ctl_tool_call_t *c, char *out_name, size_t n) {
  const char *file = tool_arg(c, "file");
  const char *out_arg = tool_arg(c, "out");
  const char *title = tool_arg(c, "title");
  char stem[200];
  if (file && file[0])
    stem_of(file, stem, sizeof(stem));
  else if (title && title[0])
    stem_of_title(title, stem, sizeof(stem));
  else
    snprintf(stem, sizeof(stem), "document");
  if (out_arg && out_arg[0]) {
    snprintf(out_name, n, "%.300s", out_arg);
    size_t ol = strlen(out_name);
    if (ol && (out_name[ol - 1] == '/' || out_name[ol - 1] == '\\') &&
        ol + strlen(stem) + 6 < n)
      snprintf(out_name + ol, n - ol, "%s.html", stem);
  } else {
    snprintf(out_name, n, "%s.html", stem);
  }
}

/* What the window shows before the write is approved. */
int cst_preview_md_to_html(const cst_tool_ctx_t *ctx, const ctl_tool_call_t *c,
                           char **out) {
  (void)ctx;
  const char *md = tool_arg(c, "markdown");
  const char *file = tool_arg(c, "file");
  if ((!md || !md[0]) && (!file || !file[0])) {
    *out = strdup("error: give markdown, or a .md file to render");
    return -1;
  }
  char out_name[400];
  md_out_name(c, out_name, sizeof(out_name));
  cst_buf_t b = {0};
  if (file && file[0])
    cst_buf_printf(&b, "render %.200s → %.200s\n", file, out_name);
  else
    cst_buf_printf(&b, "render %zu bytes of markdown → %.200s\n",
                   md ? strlen(md) : 0, out_name);
  *out = cst_buf_steal(&b);
  return 0;
}

int cst_tool_md_to_html(const cst_tool_ctx_t *ctx, const ctl_tool_call_t *c,
                        cst_tool_result_t *r) {
  const char *md = tool_arg(c, "markdown");
  const char *file = tool_arg(c, "file");
  if ((!md || !md[0]) && (!file || !file[0]))
    return tool_set_result(r, "error: give markdown, or a .md file to render",
                           1);
  const char *text = md;
  size_t len = md ? strlen(md) : 0;
  char *owned = NULL;
  if (file && file[0]) {
    char src[OS_PATH_MAX];
    if (cst_sandbox_resolve(ctx->workspace_root, file, src, sizeof(src)) != 0)
      return tool_set_resultf(r, 1,
                              "error: path outside workspace or missing "
                              "parent: %.200s",
                              file);
    size_t n = 0;
    owned = tool_slurp(src, &n, MD_TO_HTML_MAX);
    if (!owned)
      return tool_set_resultf(r, 1,
                              "error: cannot read %.200s (or it is larger than "
                              "%d MB)",
                              file, MD_TO_HTML_MAX >> 20);
    text = owned;
    len = n;
  }

  char title[256], out_name[400];
  const char *given = tool_arg(c, "title");
  snprintf(title, sizeof(title), "%.200s", given && given[0] ? given : "Document");
  md_out_name(c, out_name, sizeof(out_name));

  char *html = cst_web_md_document(text, len, title);
  free(owned);
  if (!html)
    return tool_set_result(r, "error: cannot render the markdown", 1);
  tool_ensure_parent_dirs(ctx->workspace_root, out_name);
  char full[OS_PATH_MAX];
  if (cst_sandbox_resolve(ctx->workspace_root, out_name, full, sizeof(full)) !=
      0) {
    free(html);
    return tool_set_resultf(r, 1, "error: path outside workspace: %.200s",
                            out_name);
  }
  size_t bytes = strlen(html);
  int wrote = tool_write_whole(full, html, bytes);
  free(html);
  if (wrote != 0)
    return tool_set_resultf(r, 1, "error: cannot write %.200s", out_name);

  const char *session = ctx->agent ? cst_agent_session_path(ctx->agent) : NULL;
  cst_attachment_t *a = NULL;
  char err[256] = "";
  if (session && session[0])
    a = cst_attach_import_artifact(session, full, NULL, err, sizeof(err));

  cst_buf_t out = {0};
  cst_buf_printf(&out,
                 "wrote %.200s (%zu bytes): one HTML file with everything "
                 "inside it, ready to open in a browser\n",
                 out_name, bytes);
  if (a)
    cst_buf_puts(&out,
                 "(the user sees it as a file card they can open or share; "
                 "you will not be shown its bytes again)\n");
  else if (err[0])
    cst_buf_printf(&out, "(the file is in the workspace; handing it over "
                         "failed: %.150s)\n",
                   err);
  int rc = tool_take_buf(r, &out, 0);
  if (a) {
    r->attach = a;
    r->nattach = 1;
  }
  return rc;
}

/* --- download ------------------------------------------------------------ */

#define DOWNLOAD_MAX (128 * 1024 * 1024)
#define DOWNLOAD_TIMEOUT 300L

typedef struct {
  FILE *fp;
  size_t written;
  size_t cap;
  int overflow;
  int failed;
} dl_t;

static size_t on_file(char *p, size_t sz, size_t n, void *ud) {
  dl_t *d = ud;
  size_t len = sz * n;
  if (d->written + len > d->cap) {
    d->overflow = 1;
    return 0; /* abort: the cap is what keeps a wrong URL from filling a disk */
  }
  if (fwrite(p, 1, len, d->fp) != len) {
    d->failed = 1;
    return 0;
  }
  d->written += len;
  return len;
}

int cst_tool_download(const cst_tool_ctx_t *ctx, const ctl_tool_call_t *c,
                      cst_tool_result_t *r) {
  const char *url = tool_arg(c, "url");
  if (!url || !url[0])
    return tool_set_result(r, "error: missing url", 1);
  while (*url == ' ')
    url++;
  if (strncasecmp(url, "http://", 7) != 0 &&
      strncasecmp(url, "https://", 8) != 0)
    return tool_set_result(r, "error: only http:// and https:// URLs", 1);
  long cap = tool_arg_long(c, "max_bytes", DOWNLOAD_MAX);
  if (cap <= 0 || cap > DOWNLOAD_MAX)
    cap = DOWNLOAD_MAX;

  const char *path = tool_arg(c, "path");
  char rel[400];
  if (path && path[0])
    snprintf(rel, sizeof(rel), "%s", path);
  else
    name_from_url(url, rel, sizeof(rel));
  size_t rl = strlen(rel);
  if (rl && (rel[rl - 1] == '/' || rel[rl - 1] == '\\')) {
    /* the caller named the directory: the URL names the file */
    char name[200];
    name_from_url(url, name, sizeof(name));
    snprintf(rel + rl, sizeof(rel) - rl, "%s", name);
  }
  tool_ensure_parent_dirs(ctx->workspace_root, rel);
  char full[OS_PATH_MAX];
  if (cst_sandbox_resolve(ctx->workspace_root, rel, full, sizeof(full)) != 0)
    return tool_set_resultf(r, 1,
                            "error: path outside workspace or missing parent: "
                            "%.200s",
                            rel);
  FILE *fp = fopen(full, "wb");
  if (!fp)
    return tool_set_resultf(r, 1, "error: cannot write %.200s", rel);

  CURL *curl = curl_easy_init();
  if (!curl) {
    fclose(fp);
    remove(full);
    return tool_set_result(r, "error: curl init failed", 1);
  }
  dl_t d;
  memset(&d, 0, sizeof(d));
  d.fp = fp;
  d.cap = (size_t)cap;
  char errbuf[CURL_ERROR_SIZE];
  errbuf[0] = 0;
  cst_curl_apply_ca(curl);
  curl_easy_setopt(curl, CURLOPT_URL, url);
  curl_easy_setopt(curl, CURLOPT_FOLLOWLOCATION, 1L);
  curl_easy_setopt(curl, CURLOPT_MAXREDIRS, 5L);
#if LIBCURL_VERSION_NUM >= 0x075500
  curl_easy_setopt(curl, CURLOPT_PROTOCOLS_STR, "http,https");
  curl_easy_setopt(curl, CURLOPT_REDIR_PROTOCOLS_STR, "http,https");
#else
  curl_easy_setopt(curl, CURLOPT_PROTOCOLS,
                   (long)(CURLPROTO_HTTP | CURLPROTO_HTTPS));
  curl_easy_setopt(curl, CURLOPT_REDIR_PROTOCOLS,
                   (long)(CURLPROTO_HTTP | CURLPROTO_HTTPS));
#endif
  curl_easy_setopt(curl, CURLOPT_CONNECTTIMEOUT, 15L);
  curl_easy_setopt(curl, CURLOPT_TIMEOUT, DOWNLOAD_TIMEOUT);
  curl_easy_setopt(curl, CURLOPT_NOSIGNAL, 1L);
  curl_easy_setopt(curl, CURLOPT_USERAGENT, "cst-agent/1.0 (+download)");
  curl_easy_setopt(curl, CURLOPT_WRITEFUNCTION, on_file);
  curl_easy_setopt(curl, CURLOPT_WRITEDATA, &d);
  curl_easy_setopt(curl, CURLOPT_ERRORBUFFER, errbuf);
  CURLcode rc = curl_easy_perform(curl);
  long http = 0;
  char *ctype = NULL;
  char *final_url = NULL;
  curl_easy_getinfo(curl, CURLINFO_RESPONSE_CODE, &http);
  curl_easy_getinfo(curl, CURLINFO_CONTENT_TYPE, &ctype);
  curl_easy_getinfo(curl, CURLINFO_EFFECTIVE_URL, &final_url);
  int closed = fclose(fp) != 0;
  curl_easy_cleanup(curl);
  if (rc != CURLE_OK || d.failed || closed || http >= 400) {
    remove(full); /* half a file is worse than none */
    if (d.overflow)
      return tool_set_resultf(r, 1,
                              "error: %.200s is larger than %ld bytes and "
                              "nothing was saved (raise max_bytes, up to %d "
                              "MB)",
                              url, cap, DOWNLOAD_MAX >> 20);
    return tool_set_resultf(r, 1, "error: download failed: HTTP %ld %.200s",
                            http, errbuf[0] ? errbuf : curl_easy_strerror(rc));
  }
  cst_buf_t b = {0};
  cst_buf_printf(&b, "saved %.200s (%s, %zu bytes) from HTTP %ld %.200s\n", rel,
                 ctype ? ctype : "application/octet-stream", d.written, http,
                 final_url ? final_url : url);
  cst_buf_puts(&b,
               "(the bytes are on disk, not in this answer; read_file or "
               "read_pdf it when you need to look inside)\n");
  return tool_take_buf(r, &b, 0);
}