| 1 | package bot |
| 2 | |
| 3 | // Feed discovery — the builder's answer to "regular people don't know RSS". |
| 4 | // A member names a publication ("the Irish Times"); the builder translates |
| 5 | // well-known outlets from its own knowledge, and for the long tail calls |
| 6 | // discover_feeds: fetch the site the MEMBER just named and read the feed |
| 7 | // URLs its HTML head advertises (<link rel="alternate" type="…rss/atom…">). |
| 8 | // The drafting table's dry run then proves the chosen feeds actually work |
| 9 | // before anything is created — a wrong guess fails visibly, not silently. |
| 10 | // |
| 11 | // Bounds (this fetch is NOT allowlist-gated — the member named the site, in |
| 12 | // a DM, to the builder; that ask is the consent): https only, public |
| 13 | // addresses only (the same dialer guard as task fetches), HTML head sized |
| 14 | // (256 KiB), one host, no cross-host redirects. |
| 15 | |
| 16 | import ( |
| 17 | "fmt" |
| 18 | "io" |
| 19 | "net/http" |
| 20 | "net/url" |
| 21 | "regexp" |
| 22 | "strings" |
| 23 | ) |
| 24 | |
| 25 | const maxDiscoverBytes = 256 * 1024 |
| 26 | |
| 27 | // discoverFeeds fetches a site's homepage and returns the feed URLs it |
| 28 | // advertises, absolute, deduped, capped at 8. |
| 29 | func discoverFeeds(raw string) ([]string, error) { |
| 30 | raw = strings.TrimSpace(raw) |
| 31 | if raw == "" { |
| 32 | return nil, fmt.Errorf("no site given") |
| 33 | } |
| 34 | if !strings.Contains(raw, "://") { |
| 35 | raw = "https://" + raw |
| 36 | } |
| 37 | u, err := url.Parse(raw) |
| 38 | if err != nil || u.Host == "" { |
| 39 | return nil, fmt.Errorf("that doesn't look like a site address") |
| 40 | } |
| 41 | if u.Scheme != "https" && !allowPrivateFetch() { |
| 42 | return nil, fmt.Errorf("only https sites can be read") |
| 43 | } |
| 44 | client := fetchClient([]string{}) // the SSRF-guarded dialer; no allowlist here |
| 45 | client.CheckRedirect = func(req *http.Request, via []*http.Request) error { |
| 46 | if len(via) >= 3 { |
| 47 | return fmt.Errorf("too many redirects") |
| 48 | } |
| 49 | // The www dance is the same site: members type "bbc.com", the apex |
| 50 | // redirects to www (or vice versa). Anything further off-host is a |
| 51 | // real departure and stays blocked. |
| 52 | if !sameSiteHost(req.URL.Host, u.Host) && !allowPrivateFetch() { |
| 53 | return fmt.Errorf("site redirected off-host") |
| 54 | } |
| 55 | return nil |
| 56 | } |
| 57 | req, err := http.NewRequest(http.MethodGet, u.String(), nil) |
| 58 | if err != nil { |
| 59 | return nil, err |
| 60 | } |
| 61 | req.Header.Set("User-Agent", "lchat-bot/1.0 (+https://elevenmessenger.com)") |
| 62 | req.Header.Set("Accept", "text/html") |
| 63 | resp, err := client.Do(req) |
| 64 | if err != nil { |
| 65 | return nil, fmt.Errorf("couldn't reach %s: %w", u.Host, err) |
| 66 | } |
| 67 | defer resp.Body.Close() |
| 68 | if resp.StatusCode != http.StatusOK { |
| 69 | return nil, fmt.Errorf("%s answered with status %d", u.Host, resp.StatusCode) |
| 70 | } |
| 71 | body, err := io.ReadAll(io.LimitReader(resp.Body, maxDiscoverBytes)) |
| 72 | if err != nil { |
| 73 | return nil, err |
| 74 | } |
| 75 | feeds := feedLinksFromHTML(string(body), resp.Request.URL) |
| 76 | if len(feeds) == 0 { |
| 77 | return nil, fmt.Errorf("%s doesn't advertise a feed on its homepage", u.Host) |
| 78 | } |
| 79 | return feeds, nil |
| 80 | } |
| 81 | |
| 82 | // sameSiteHost treats a host and its www. variant as one site (either |
| 83 | // direction), for the redirect guard above. Pure; pinned by unit test. |
| 84 | func sameSiteHost(a, b string) bool { |
| 85 | return a == b || a == "www."+b || b == "www."+a |
| 86 | } |
| 87 | |
| 88 | // linkTagRE matches <link …> tags; attribute order varies wildly in the |
| 89 | // field, so attributes are re-scanned per tag rather than positionally. |
| 90 | var linkTagRE = regexp.MustCompile(`(?is)<link\b[^>]*>`) |
| 91 | var attrRE = regexp.MustCompile(`(?is)([a-z-]+)\s*=\s*("([^"]*)"|'([^']*)'|([^\s"'>]+))`) |
| 92 | |
| 93 | // feedLinksFromHTML pulls rel="alternate" rss/atom link hrefs out of an HTML |
| 94 | // document, resolved against base. Pure — pinned by unit tests. |
| 95 | func feedLinksFromHTML(html string, base *url.URL) []string { |
| 96 | var out []string |
| 97 | seen := map[string]bool{} |
| 98 | for _, tag := range linkTagRE.FindAllString(html, 200) { |
| 99 | attrs := map[string]string{} |
| 100 | for _, m := range attrRE.FindAllStringSubmatch(tag, -1) { |
| 101 | val := m[3] |
| 102 | if val == "" { |
| 103 | val = m[4] |
| 104 | } |
| 105 | if val == "" { |
| 106 | val = m[5] |
| 107 | } |
| 108 | attrs[strings.ToLower(m[1])] = val |
| 109 | } |
| 110 | rel := strings.ToLower(attrs["rel"]) |
| 111 | typ := strings.ToLower(attrs["type"]) |
| 112 | href := strings.TrimSpace(attrs["href"]) |
| 113 | if href == "" || !strings.Contains(rel, "alternate") { |
| 114 | continue |
| 115 | } |
| 116 | if !strings.Contains(typ, "rss") && !strings.Contains(typ, "atom") { |
| 117 | continue |
| 118 | } |
| 119 | ref, err := url.Parse(href) |
| 120 | if err != nil { |
| 121 | continue |
| 122 | } |
| 123 | abs := base.ResolveReference(ref).String() |
| 124 | if seen[abs] { |
| 125 | continue |
| 126 | } |
| 127 | seen[abs] = true |
| 128 | out = append(out, abs) |
| 129 | if len(out) == 8 { |
| 130 | break |
| 131 | } |
| 132 | } |
| 133 | return out |
| 134 | } |
| 135 | |
| 136 | // discoverFeedsTool is the builder's information tool (a RESULT tool — the |
| 137 | // brain round-trips its output, unlike chip tools; see modelReply's loop). |
| 138 | func discoverFeedsTool() anthTool { |
| 139 | return anthTool{ |
| 140 | Name: "discover_feeds", |
| 141 | Description: "Find the RSS/Atom feed URLs a website advertises. Give it the site a member " + |
| 142 | "named (a domain or homepage URL, e.g. irishtimes.com) and it returns the feed URLs from " + |
| 143 | "the site's own HTML. Use this whenever you aren't CERTAIN of an outlet's feed URL — " + |
| 144 | "never ask a member for a feed URL, and never guess one you can't vouch for.", |
| 145 | InputSchema: map[string]any{ |
| 146 | "type": "object", |
| 147 | "properties": map[string]any{"site": map[string]any{"type": "string", "description": "Domain or homepage URL"}}, |
| 148 | "required": []any{"site"}, |
| 149 | }, |
| 150 | } |
| 151 | } |
| 152 | |
| 153 | // runDiscoverFeeds executes the tool, formatting for a tool_result. |
| 154 | func runDiscoverFeeds(input map[string]any) string { |
| 155 | site, _ := input["site"].(string) |
| 156 | feeds, err := discoverFeeds(site) |
| 157 | if err != nil { |
| 158 | return "discovery failed: " + err.Error() |
| 159 | } |
| 160 | return "feeds advertised by the site:\n" + strings.Join(feeds, "\n") |
| 161 | } |
| 162 | |
amadan