a digital entity named phi that roams bsky
phi.zzstoatzz.io
fix: resolve facet link URIs so phi sees full URLs, not truncated display text
bluesky truncates long URLs in post display text (e.g. "example.com/long...")
but stores the actual URI in the facet. phi was only reading post.record.text,
so it never saw the real link — causing it to guess slugs in a loop.
add resolve_facet_links() that splices facet URIs back into the text at
their byte offsets. used in both _handle_post (mention text) and
describe_post (thread context).
reverts all agent.py changes from v0.0.13/v0.0.14 — those were treating
downstream symptoms, not the root cause.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
+174
-24
+3
-19
src/bot/agent.py
+3
-19
src/bot/agent.py
···
97
97
"""Build operational instructions with the current owner handle interpolated."""
98
98
return f"""
99
99
indicate your response action via the structured output — do not use atproto tools to post, like, or repost directly.
100
-
when sharing URLs you found yourself (not ones the user gave you), you can verify them with check_urls. always include https://. never call check_urls more than once per URL — accept the result.
100
+
when sharing URLs, verify them with check_urls first and always include https://.
101
101
102
102
you receive all notification types — mentions, replies, quotes, likes, reposts, and follows.
103
103
for mentions, replies, and quotes: someone is talking to you or about you. respond if you have something to say.
···
142
142
only use this during daily reflection, or when someone explicitly asks about infrastructure, services, or uptime of specific apps.
143
143
if something is down, post about it and tag @{settings.owner_handle}.
144
144
145
-
IMPORTANT: you have a limited tool call budget per message. do not call the same tool repeatedly with the same or similar arguments — accept the result you get. never paginate through list_records repeatedly. if you need more data than one call returns, work with what you have. if you cannot fulfill a request with the tools you have (e.g. reading a web page), say so instead of looping.
145
+
IMPORTANT: never paginate through list_records repeatedly. if you need more data than one call returns, work with what you have. endless pagination wastes your request budget and produces no response.
146
146
""".strip()
147
147
148
148
···
638
638
except Exception as e:
639
639
return f"failed to post: {e}"
640
640
641
-
self._checked_urls: set[str] = set()
642
-
643
641
@self.agent.tool
644
642
async def check_urls(ctx: RunContext[PhiDeps], urls: list[str]) -> str:
645
-
"""Check whether URLs are reachable (HEAD request only — cannot read page content). If a URL 404s, do not guess alternative slugs — tell the user you can't find it."""
646
-
647
-
# if we've already checked many URLs, the model is probably guessing slugs
648
-
new_urls = [u for u in urls if u not in self._checked_urls]
649
-
if not new_urls and self._checked_urls:
650
-
return "already checked these URLs. stop guessing and respond with what you know."
651
-
if len(self._checked_urls) > 10:
652
-
return (
653
-
f"you've already checked {len(self._checked_urls)} URLs in this conversation. "
654
-
"stop guessing slugs — you cannot browse the web. tell the user you "
655
-
"couldn't find the exact page and share what you found instead."
656
-
)
643
+
"""Check whether URLs are reachable. Use this before sharing links to verify they actually work. Accepts full URLs (https://...) or bare domains (example.com/path)."""
657
644
658
645
async def _check(client: httpx.AsyncClient, url: str) -> str:
659
646
if not url.startswith(("http://", "https://")):
660
647
url = f"https://{url}"
661
-
self._checked_urls.add(url)
662
648
try:
663
649
hostname = urlparse(url).hostname
664
650
if not hostname:
···
848
834
) -> Response:
849
835
"""Process a mention with structured memory context."""
850
836
logger.info(f"processing mention from @{author_handle}: {mention_text[:80]}")
851
-
self._checked_urls.clear()
852
837
853
838
deps = PhiDeps(
854
839
author_handle=author_handle,
···
892
877
async def process_reflection(self, last_post_text: str | None = None) -> Response:
893
878
"""Generate a daily reflection post from recent memory."""
894
879
logger.info("processing daily reflection")
895
-
self._checked_urls.clear()
896
880
897
881
# Pre-fetch context that doesn't benefit from semantic search against the prompt
898
882
recent_activity = ""
+7
-2
src/bot/services/message_handler.py
+7
-2
src/bot/services/message_handler.py
···
11
11
from bot.agent import PhiAgent
12
12
from bot.core.atproto_client import BotClient
13
13
from bot.status import bot_status
14
-
from bot.utils.thread import build_thread_context, describe_embed, extract_image_urls
14
+
from bot.utils.thread import (
15
+
build_thread_context,
16
+
describe_embed,
17
+
extract_image_urls,
18
+
resolve_facet_links,
19
+
)
15
20
16
21
logger = logging.getLogger("bot.handler")
17
22
···
135
140
return
136
141
137
142
post = posts.posts[0]
138
-
mention_text = post.record.text
143
+
mention_text = resolve_facet_links(post.record)
139
144
author_handle = post.author.handle
140
145
141
146
# Include embed content (images, links, quote posts) in the mention
+52
-3
src/bot/utils/thread.py
+52
-3
src/bot/utils/thread.py
···
3
3
from collections.abc import Callable
4
4
5
5
6
+
def resolve_facet_links(record) -> str:
7
+
"""Return post text with truncated link display text replaced by actual URIs from facets.
8
+
9
+
Bluesky truncates long URLs in the display text (e.g. "example.com/long-path..."
10
+
but stores the full URI in the facet. This walks facets right-to-left and splices
11
+
the real URI back into the text so downstream consumers see the full link.
12
+
"""
13
+
text = getattr(record, "text", "") or ""
14
+
facets = getattr(record, "facets", None)
15
+
if not facets:
16
+
return text
17
+
18
+
# collect link facets with byte ranges
19
+
link_facets = []
20
+
for facet in facets:
21
+
index = getattr(facet, "index", None)
22
+
features = getattr(facet, "features", None) or []
23
+
if not index:
24
+
continue
25
+
for feature in features:
26
+
py_type = getattr(feature, "py_type", "")
27
+
if "link" in py_type:
28
+
uri = getattr(feature, "uri", "")
29
+
if uri:
30
+
link_facets.append(
31
+
(
32
+
getattr(index, "byte_start", 0),
33
+
getattr(index, "byte_end", 0),
34
+
uri,
35
+
)
36
+
)
37
+
38
+
if not link_facets:
39
+
return text
40
+
41
+
# sort by byte_start descending so replacements don't shift earlier offsets
42
+
link_facets.sort(key=lambda x: x[0], reverse=True)
43
+
44
+
encoded = text.encode("utf-8")
45
+
for start, end, uri in link_facets:
46
+
encoded = encoded[:start] + uri.encode("utf-8") + encoded[end:]
47
+
48
+
return encoded.decode("utf-8")
49
+
50
+
6
51
def describe_embed(embed) -> str | None:
7
52
"""Extract a human-readable description from a post embed.
8
53
···
110
155
def describe_post(post) -> str:
111
156
"""Build a full text representation of a post including embeds."""
112
157
handle = post.author.handle
113
-
text = post.record.text if hasattr(post.record, "text") else ""
158
+
text = resolve_facet_links(post.record) if hasattr(post.record, "text") else ""
114
159
115
160
# Check for embeds on the post view (post.embed) or record (post.record.embed)
116
161
embed_desc = None
···
120
165
embed_desc = describe_embed(post.record.embed)
121
166
122
167
if embed_desc:
123
-
return f"@{handle}: {text}\n {embed_desc}" if text else f"@{handle}: {embed_desc}"
168
+
return (
169
+
f"@{handle}: {text}\n {embed_desc}" if text else f"@{handle}: {embed_desc}"
170
+
)
124
171
return f"@{handle}: {text}" if text else f"@{handle}: [no text]"
125
172
126
173
···
151
198
152
199
# Traverse parent chain (moving up the thread)
153
200
if include_parent and hasattr(thread_node, "parent") and thread_node.parent:
154
-
traverse_thread(thread_node.parent, visit, include_parent=True, include_replies=False)
201
+
traverse_thread(
202
+
thread_node.parent, visit, include_parent=True, include_replies=False
203
+
)
155
204
156
205
# Traverse replies (moving down the thread)
157
206
if include_replies and hasattr(thread_node, "replies") and thread_node.replies:
+112
tests/test_resolve_facets.py
+112
tests/test_resolve_facets.py
···
1
+
"""Regression tests for resolve_facet_links — ensures phi sees full URLs from facets."""
2
+
3
+
from types import SimpleNamespace
4
+
5
+
from bot.utils.thread import resolve_facet_links
6
+
7
+
8
+
def _make_record(text, facets=None):
9
+
return SimpleNamespace(text=text, facets=facets)
10
+
11
+
12
+
def _make_facet(byte_start, byte_end, uri):
13
+
return SimpleNamespace(
14
+
index=SimpleNamespace(byte_start=byte_start, byte_end=byte_end),
15
+
features=[SimpleNamespace(py_type="app.bsky.richtext.facet#link", uri=uri)],
16
+
)
17
+
18
+
19
+
def test_no_facets():
20
+
record = _make_record("hello world")
21
+
assert resolve_facet_links(record) == "hello world"
22
+
23
+
24
+
def test_none_facets():
25
+
record = _make_record("hello world", facets=None)
26
+
assert resolve_facet_links(record) == "hello world"
27
+
28
+
29
+
def test_truncated_url_replaced():
30
+
"""The exact bug from trace 019d5004 — bluesky truncated the URL in display text."""
31
+
text = (
32
+
"cool. onto something else\n\n"
33
+
"can you read this: www.letta.com/blog/context...\n\n"
34
+
"and tell me whether there's anything interesting as far as your M.O.?"
35
+
)
36
+
# byte offsets for "www.letta.com/blog/context..." in the text
37
+
encoded = text.encode("utf-8")
38
+
start = encoded.index(b"www.letta.com/blog/context...")
39
+
end = start + len(b"www.letta.com/blog/context...")
40
+
41
+
record = _make_record(
42
+
text,
43
+
facets=[
44
+
_make_facet(start, end, "https://www.letta.com/blog/context-constitution")
45
+
],
46
+
)
47
+
48
+
result = resolve_facet_links(record)
49
+
assert "https://www.letta.com/blog/context-constitution" in result
50
+
assert "context..." not in result
51
+
52
+
53
+
def test_multiple_links():
54
+
text = "check out link1... and link2..."
55
+
encoded = text.encode("utf-8")
56
+
s1 = encoded.index(b"link1...")
57
+
e1 = s1 + len(b"link1...")
58
+
s2 = encoded.index(b"link2...")
59
+
e2 = s2 + len(b"link2...")
60
+
61
+
record = _make_record(
62
+
text,
63
+
facets=[
64
+
_make_facet(s1, e1, "https://example.com/link1-full"),
65
+
_make_facet(s2, e2, "https://example.com/link2-full"),
66
+
],
67
+
)
68
+
69
+
result = resolve_facet_links(record)
70
+
assert "https://example.com/link1-full" in result
71
+
assert "https://example.com/link2-full" in result
72
+
assert "link1..." not in result
73
+
assert "link2..." not in result
74
+
75
+
76
+
def test_mention_facet_ignored():
77
+
"""Mention facets should not affect the text."""
78
+
text = "hey @someone check this"
79
+
record = _make_record(
80
+
text,
81
+
facets=[
82
+
SimpleNamespace(
83
+
index=SimpleNamespace(byte_start=4, byte_end=12),
84
+
features=[
85
+
SimpleNamespace(
86
+
py_type="app.bsky.richtext.facet#mention",
87
+
did="did:plc:abc",
88
+
)
89
+
],
90
+
)
91
+
],
92
+
)
93
+
assert resolve_facet_links(record) == text
94
+
95
+
96
+
def test_unicode_text_byte_offsets():
97
+
"""Facet byte offsets are in UTF-8 bytes, not characters."""
98
+
# emoji is 4 bytes in UTF-8
99
+
text = "\U0001f600 see link..."
100
+
encoded = text.encode("utf-8")
101
+
start = encoded.index(b"link...")
102
+
end = start + len(b"link...")
103
+
104
+
record = _make_record(
105
+
text,
106
+
facets=[_make_facet(start, end, "https://example.com/full-link")],
107
+
)
108
+
109
+
result = resolve_facet_links(record)
110
+
assert "https://example.com/full-link" in result
111
+
assert "link..." not in result
112
+
assert result.startswith("\U0001f600")