markedet commited on
Commit
a101db0
·
verified ·
1 Parent(s): e34dc18

Update van markedet

Browse files
README.md CHANGED
@@ -44,13 +44,13 @@ Three parts:
44
 
45
  | Part | Records | What it is |
46
  |---|---|---|
47
- | `pitfall-cards/` | 361 | Situation, symptom, exact error strings, cause, fix, snippet, verify step. |
48
  | `deprecations/` | 326 | Deprecated or retiring models, API versions and endpoints, with replacement and shutdown date. |
49
  | `needs-a-human-map/` | 34 | For common services, which documented authentication routes work without a browser, and what a person must do once. |
50
 
51
  ## How an agent should query this
52
 
53
- One read gives the overview: open `llms.txt`. Then search `index.jsonl` (721 lines, one per record, fields `id, part, file, title, tags, error_strings, source, checked_on` and, when a pitfall card revises or is revised, `revision_of` and/or `superseded_by`) and open only the file it points to. Cards are short and self-contained, so you never need to read the full files.
54
 
55
  ```
56
  # search by the exact error text you see
@@ -133,12 +133,12 @@ Search `needs-a-human-map/auth_map.jsonl`: 34 documented routes and the one step
133
 
134
  ## Pitfall cards
135
 
136
- Fixed fields on every card: `id, title, topic, tags, situation, symptom, error_strings, cause, fix, snippet, verify, source, source_excerpt, checked_on, license`. Optional fields on some later cards: `subject` (232 cards), `revised_on` and `revision_note` (26 cards each), `source_2` (8 cards) and `extra_sources` (1 card), and `checked_model` and `checked_tool_version` (1 card each) and `revision_of` (25 cards). When the check behind a card called a hosted model, `checked_model` gives the exact model ID taken from the logs of that run and `checked_tool_version` the client version used; `checked_on` stays the date of the run. `revision_of` names an earlier card that the entry revises; the earlier card stays as it was (PC-177 revises PC-137 and adds the model ID; later revisions include PC-179 of PC-163, PC-180 of PC-175, PC-182 of PC-146, PC-183 of PC-083, PC-184 of PC-085, PC-185 of PC-090, PC-186 of PC-091, PC-187 of PC-096, PC-231 of PC-218, PC-232 of PC-219, and PC-233 to PC-242 of ten Agent Commons cards (PC-082, PC-084, PC-087, PC-089, PC-092, PC-093, PC-094, PC-099, PC-102, PC-122), each now citing the post permalink; PC-243 of PC-221, PC-244 of PC-222, PC-245 of PC-223 and PC-247 of PC-227).
137
 
138
  - `error_strings` holds status codes and message texts literally. Message texts are only included when they appear on the cited page; standard HTTP status lines are taken from RFC 9110 and RFC 6585.
139
  - `snippet` is a command or short code to copy; `verify` says how to confirm the fix worked.
140
  - `source_excerpt`: for cards that cite an official page, standard or paper, a verbatim sentence from that source that supports the card, matched against the fetched text automatically. For cards based on agent field reports or our own checks, it is a short paraphrased statement of what was reported or observed.
141
- - Topics (87, in order of their lowest card id): Misleading status codes (13), Redirects (3), Silent truncation and pagination (15), Rate limits and Retry-After (7), OAuth tokens and device flow (5), Headless and non-interactive CLI use (10), Context-window bloat from MCP tools (6), Instruction files per tool (8), Remote MCP and OAuth per client (13), Prompt injection and untrusted content (5), Multi-agent coordination (2), Agent economy concepts (3), LLM API integration (3), Local stdio MCP setup (4), MCP tool calls (2), Misleading API responses (13), Code analysis (1), MCP list pagination (1), Network path (3), Data freshness (6), Board visibility (3), Aggregation (3), Verification (60), Identity (2), Provenance (4), Reliability (6), Async control (2), Dataset diff (3), Staleness (4), Custody (3), Observability (3), Session persistence (2), Measurement (15), Logging (3), TLS trust store (2), Quoting terms (2), Cache completeness (2), Append only vs editable (1), Green eval vs tool policy (2), Empty slot under unreadable code (2), Crawl brief per board marks (2), CSV duplicate header silent collapse (1), Hash normalization vs loss (1), Sample schema invented fields (1), Dual clock event skew (1), Parked future orphan after timeout (1), Acceptance criteria drift (1), Key rotation orphaned graph (1), Retirement grep skim (1), Artifact vs running instance (1), Amendment accession dedup (1), Askrate heuristic precision (1), Pin end anchor silent fallback (1), Volatile fact restatement drift (1), Hashing (12), Data tables (2), ID semantics (1), Deploy verification (2), Secrets (2), Exit status (2), Permissions (5), Dependencies (1), Agent discovery (4), HTTP clients (1), Git (11), Scheduling (6), Caching (2), Citations (2), HTTP (8), Transactions (1), Authentication (2), Encoding (2), Data modeling (6), Write fidelity (2), Inbox (4), Pricing (1), Reproducibility (1), Process (5), Registration (2), Signing (2), Payments (4), MCP (1), Agent hooks (1), Agent board API (1), Document export (1), Sockets (1), Porting (1).
142
  - A readable version grouped by topic: `pitfall-cards/cards.md`.
143
 
144
  ## Deprecations feed
@@ -198,7 +198,7 @@ When you report a check, say which files or cards you covered, for example "3 of
198
  ## Scope and limits
199
 
200
  - Official documentation, standards and, for the prompt-injection, multi-agent and agent-economy cards only, named research papers by their authors. The LLM API integration card cites a public issue in the provider's own repository.
201
- - 262 of the 361 pitfall cards (all from PC-081 on) also rest on field reports by agents on public agent forums, paraphrased and without agent names, and/or on our own local runs and read-only checks; the `source` field says which. 60 of them have no URL in `source`. Treat these cards as reported practice, check the `verify` step yourself, and prefer the official source where one is cited. Documentation and standards: provider docs, tool docs, and RFCs and specs (RFC 9110, 6749, 8628, 8259, 9728, the WHATWG Fetch standard, the GraphQL over HTTP spec, the MCP specification). Percentages in the research-paper cards are findings that the paper reports from other studies and have not been re-checked here.
202
  - Deprecations and the needs-a-human map are a snapshot of 2026-10-29; pitfall cards were checked between 2026-10-29 and 2026-11-16. Check `checked_on` and the source before relying on a record.
203
  - Not exhaustive. Some vendors are not covered because their pages could not be fetched or checked automatically.
204
 
 
44
 
45
  | Part | Records | What it is |
46
  |---|---|---|
47
+ | `pitfall-cards/` | 370 | Situation, symptom, exact error strings, cause, fix, snippet, verify step. |
48
  | `deprecations/` | 326 | Deprecated or retiring models, API versions and endpoints, with replacement and shutdown date. |
49
  | `needs-a-human-map/` | 34 | For common services, which documented authentication routes work without a browser, and what a person must do once. |
50
 
51
  ## How an agent should query this
52
 
53
+ One read gives the overview: open `llms.txt`. Then search `index.jsonl` (730 lines, one per record, fields `id, part, file, title, tags, error_strings, source, checked_on` and, when a pitfall card revises or is revised, `revision_of` and/or `superseded_by`) and open only the file it points to. Cards are short and self-contained, so you never need to read the full files.
54
 
55
  ```
56
  # search by the exact error text you see
 
133
 
134
  ## Pitfall cards
135
 
136
+ Fixed fields on every card: `id, title, topic, tags, situation, symptom, error_strings, cause, fix, snippet, verify, source, source_excerpt, checked_on, license`. Optional fields on some later cards: `subject` (241 cards), `revised_on` and `revision_note` (26 cards each), `source_2` (8 cards) and `extra_sources` (1 card), and `checked_model` and `checked_tool_version` (1 card each) and `revision_of` (25 cards). When the check behind a card called a hosted model, `checked_model` gives the exact model ID taken from the logs of that run and `checked_tool_version` the client version used; `checked_on` stays the date of the run. `revision_of` names an earlier card that the entry revises; the earlier card stays as it was (PC-177 revises PC-137 and adds the model ID; later revisions include PC-179 of PC-163, PC-180 of PC-175, PC-182 of PC-146, PC-183 of PC-083, PC-184 of PC-085, PC-185 of PC-090, PC-186 of PC-091, PC-187 of PC-096, PC-231 of PC-218, PC-232 of PC-219, and PC-233 to PC-242 of ten Agent Commons cards (PC-082, PC-084, PC-087, PC-089, PC-092, PC-093, PC-094, PC-099, PC-102, PC-122), each now citing the post permalink; PC-243 of PC-221, PC-244 of PC-222, PC-245 of PC-223 and PC-247 of PC-227).
137
 
138
  - `error_strings` holds status codes and message texts literally. Message texts are only included when they appear on the cited page; standard HTTP status lines are taken from RFC 9110 and RFC 6585.
139
  - `snippet` is a command or short code to copy; `verify` says how to confirm the fix worked.
140
  - `source_excerpt`: for cards that cite an official page, standard or paper, a verbatim sentence from that source that supports the card, matched against the fetched text automatically. For cards based on agent field reports or our own checks, it is a short paraphrased statement of what was reported or observed.
141
+ - Topics (87, in order of their lowest card id): Misleading status codes (13), Redirects (3), Silent truncation and pagination (16), Rate limits and Retry-After (7), OAuth tokens and device flow (5), Headless and non-interactive CLI use (10), Context-window bloat from MCP tools (6), Instruction files per tool (8), Remote MCP and OAuth per client (13), Prompt injection and untrusted content (5), Multi-agent coordination (2), Agent economy concepts (3), LLM API integration (3), Local stdio MCP setup (4), MCP tool calls (2), Misleading API responses (13), Code analysis (1), MCP list pagination (1), Network path (3), Data freshness (8), Board visibility (3), Aggregation (3), Verification (62), Identity (2), Provenance (4), Reliability (6), Async control (2), Dataset diff (3), Staleness (4), Custody (3), Observability (3), Session persistence (2), Measurement (15), Logging (3), TLS trust store (2), Quoting terms (2), Cache completeness (2), Append only vs editable (1), Green eval vs tool policy (2), Empty slot under unreadable code (2), Crawl brief per board marks (2), CSV duplicate header silent collapse (1), Hash normalization vs loss (1), Sample schema invented fields (1), Dual clock event skew (1), Parked future orphan after timeout (1), Acceptance criteria drift (1), Key rotation orphaned graph (1), Retirement grep skim (1), Artifact vs running instance (1), Amendment accession dedup (1), Askrate heuristic precision (1), Pin end anchor silent fallback (1), Volatile fact restatement drift (1), Hashing (12), Data tables (2), ID semantics (1), Deploy verification (2), Secrets (2), Exit status (2), Permissions (7), Dependencies (1), Agent discovery (4), HTTP clients (1), Git (11), Scheduling (6), Caching (2), Citations (2), HTTP (8), Transactions (1), Authentication (2), Encoding (2), Data modeling (6), Write fidelity (2), Inbox (4), Pricing (1), Reproducibility (1), Process (7), Registration (2), Signing (2), Payments (4), MCP (1), Agent hooks (1), Agent board API (1), Document export (1), Sockets (1), Porting (1).
142
  - A readable version grouped by topic: `pitfall-cards/cards.md`.
143
 
144
  ## Deprecations feed
 
198
  ## Scope and limits
199
 
200
  - Official documentation, standards and, for the prompt-injection, multi-agent and agent-economy cards only, named research papers by their authors. The LLM API integration card cites a public issue in the provider's own repository.
201
+ - 269 of the 370 pitfall cards (all from PC-081 on) also rest on field reports by agents on public agent forums, paraphrased and without agent names, and/or on our own local runs and read-only checks; the `source` field says which. 62 of them have no URL in `source`. Treat these cards as reported practice, check the `verify` step yourself, and prefer the official source where one is cited. Documentation and standards: provider docs, tool docs, and RFCs and specs (RFC 9110, 6749, 8628, 8259, 9728, the WHATWG Fetch standard, the GraphQL over HTTP spec, the MCP specification). Percentages in the research-paper cards are findings that the paper reports from other studies and have not been re-checked here.
202
  - Deprecations and the needs-a-human map are a snapshot of 2026-10-29; pitfall cards were checked between 2026-10-29 and 2026-11-16. Check `checked_on` and the source before relying on a record.
203
  - Not exhaustive. Some vendors are not covered because their pages could not be fetched or checked automatically.
204
 
index.jsonl CHANGED
@@ -73,7 +73,7 @@
73
  {"id": "PC-073", "part": "pitfall-cards", "file": "pitfall-cards/cards.jsonl", "title": "Identifier names in untrusted code are labels, not evidence: a model can balk at a script because of alarming names, so normalise the names and describe the behavior", "tags": ["reverse-engineering", "security-review", "javascript", "identifiers", "deobfuscation", "untrusted-content"], "error_strings": [], "source": "https://nullpt.rs/fingerprinting-bioweapons", "checked_on": "2026-10-30"}
74
  {"id": "PC-074", "part": "pitfall-cards", "file": "pitfall-cards/cards.jsonl", "title": "An empty-string nextCursor in an MCP list result is not the end: keep paging until nextCursor is missing, and never parse the cursor", "tags": ["mcp", "pagination", "nextCursor", "tools/list", "resources/list", "prompts/list"], "error_strings": [], "source": "https://modelcontextprotocol.io/specification/2026-07-28/server/utilities/pagination", "checked_on": "2026-10-30"}
75
  {"id": "PC-075", "part": "pitfall-cards", "file": "pitfall-cards/cards.jsonl", "title": "A site that resolves to 198.18.x.x is being answered by a local proxy, not by the site: report it as unreachable from your path, not as down", "tags": ["dns", "proxy", "fake-ip", "tls", "198.18.0.0/15", "reachability"], "error_strings": [], "source": "https://www.iana.org/assignments/iana-ipv4-special-registry/iana-ipv4-special-registry.xhtml", "checked_on": "2026-10-30"}
76
- {"id": "PC-076", "part": "pitfall-cards", "file": "pitfall-cards/cards.jsonl", "title": "An oracle feed value can be valid and still old: check updatedAt against the feed's heartbeat before you use it", "tags": ["oracle", "data-feeds", "heartbeat", "deviation-threshold", "updatedAt", "staleness", "chainlink", "cached", "fallback", "spot", "quote", "live price", "stale quote"], "error_strings": [], "source": "https://docs.chain.link/architecture-overview/architecture-decentralized-model", "checked_on": "2026-10-31"}
77
  {"id": "PC-077", "part": "pitfall-cards", "file": "pitfall-cards/cards.jsonl", "title": "A 200 response with a tidy list can be one bounded page: check hasMore and nextCursor before you treat it as the full history", "tags": ["pagination", "cursor", "hasMore", "nextCursor", "silent-truncation", "cap", "500 cap", "not paginated", "census", "seats"], "error_strings": ["hasMore", "nextCursor"], "source": "https://openagentforum.com/llms-full.txt", "checked_on": "2026-10-31"}
78
  {"id": "PC-078", "part": "pitfall-cards", "file": "pitfall-cards/cards.jsonl", "title": "A post on an agent board can open by id and still be missing from listings: read it back with ?agent=<your name>, and if it is not there, call it not listed, possibly hidden", "tags": ["agent-commons", "visibility", "moderation", "read-back", "silent-failure"], "error_strings": [], "source": "https://commons.dooza.ai/llms.txt", "checked_on": "2026-10-31"}
79
  {"id": "PC-079", "part": "pitfall-cards", "file": "pitfall-cards/cards.jsonl", "title": "A failed or skipped lookup must not turn into zero: give every read a three-state result and let totals refuse to add up when any input is unavailable", "tags": ["silent-failure", "aggregation", "timeouts", "partial-results", "counts", "verification", "not_attempted", "not attempted", "inconclusive", "pending", "unpriced", "two kinds of zero", "unreachable", "unreadable slot", "grade"], "error_strings": [], "source": "https://spec.graphql.org/October2021/", "checked_on": "2026-11-01"}
@@ -359,6 +359,15 @@
359
  {"id": "PC-374", "part": "pitfall-cards", "file": "pitfall-cards/cards.jsonl", "title": "A published write shape can be a subset of what the validator requires, most often on the anonymous path: when a 400 names a field the documented body does not contain, stop guessing and treat the doc as incomplete", "tags": ["api-responses", "validation", "documentation-gap", "anonymous", "http-400", "body-shape", "write-contract"], "error_strings": ["400"], "source": "https://www.velvt.ai/api/requests ; GET 200 on 2026-11-16 (19284 bytes, sha256 436a1f4f44f09f3581e6c16a6d8e3dec0ee62d226c0e38bb5c4c05fb4a3fc46e): each listed request publishes contributionContract.authentication as OPTIONAL_BUT_AUTHORITATIVE_WHEN_PROVIDED and a next[0].bodyShape with responseId, action, contributionType, research and body (no actor). research is \"Required bound result object for REPRODUCTION, FALSIFICATION, RETEST and ADOPTION; inspect research endpoint\". The list is dynamic; this hash matches later checks of the same URL on 2026-11-13 and 2026-11-14 (sha256 436a1f4f..., 19284 bytes) and differs from the hash recorded on PC-194 (f9193bff... on 2026-11-12, 12504 bytes). The core gap is unchanged: the shape still does not list an actor. One agent field report on a public agent forum, 2026-11-12, paraphrased: two anonymous POSTs of the then-documented shape both returned 400, the second naming actor.handle. We did not send those POSTs. Not measured: whether actor.handle must be a registered identity, or whether unauthenticated responses are rate limited.", "checked_on": "2026-11-16", "revision_of": "PC-194"}
360
  {"id": "PC-375", "part": "pitfall-cards", "file": "pitfall-cards/cards.jsonl", "title": "A --filter=blob:none fetch that the server ignores still exits 0: require filter on the server's fetch= capability before, run the fetch with LC_ALL=C, and treat git's ignored-filter warning or a non-zero blob count as a failure after", "tags": ["git", "partial-clone", "filter", "blob:none", "allowFilter", "protocol-v2", "stderr", "promisor", "LC_ALL", "gettext", "locale"], "error_strings": ["warning: filtering not recognized by server, ignoring", "--filter=blob:none", "uploadpack.allowFilter", "filter 'blob:none' not supported"], "source": "https://git-scm.com/docs/git-fetch ; GET 200 on 2026-11-12 (166251 bytes, sha256 1cd2f136cbd6971fbb9c3b2dd8568bf35aa4d5e028629988d9a553a629da4568, same as for PC-223): --filter=<filter-spec> uses the partial clone feature and requests that the server send a subset of reachable objects. https://git-scm.com/docs/git-config ; GET 200 on 2026-11-12 (837639 bytes, sha256 310f5954c833f3853606095b2e484ace10c241dda5f0c1911ad71e4cb4815eba, same as for PC-223): uploadpack.allowFilter makes upload-pack support partial clone and partial fetch object filtering; uploadpackfilter.<filter>.allow explicitly allows or bans one object filter. https://git-scm.com/docs/gitprotocol-v2 ; GET 200 on 2026-11-12 (84919 bytes, sha256 d423a529d9d840ed3b8f7ba65412d210bc95eaa572002c7e4abc72b408f56424): additional features are advertised as the value of the command in the capability advertisement; the filter argument may be sent if the filter feature is advertised. Own runs 2026-11-12 with git 2.47.3 on local repositories (article 52 and this card): with uploadpack.allowFilter unset, the fetch= line was \"fetch=shallow wait-for-done\", a filtered clone and a filtered unshallow printed the warning, exited 0, fetched every blob and left promisor=true and partialclonefilter=blob:none; with allowFilter true but only the tree filter allowed, the line still ended in filter and the blob:none clone failed with exit 128. One agent field report, 2026-11-05, paraphrased: a size limit was raised after a filtered fetch was not cheap. Not confirmed as a fix by a test from the agent. Additionally, 2026-11-16: the English warning string is in the local git 2.47.3 binary; git translates interface messages through gettext, so the English string is not a reliable check under another locale. Not locally proven (only locale C was present). The selftest therefore also covers a translated fixture line and a blob count.", "checked_on": "2026-11-16", "revision_of": "PC-245"}
361
  {"id": "PC-376", "part": "pitfall-cards", "file": "pitfall-cards/cards.jsonl", "title": "A matching body digest does not show your request arrived as sent: Content-Digest covers content, not fields, so record request-header names and a hash of the values (never the secrets) and compare both", "tags": ["http", "content-digest", "rfc9530", "rfc9110", "outbound-proxy", "proxy", "request-headers", "provenance", "silent-failure"], "error_strings": ["Content-Digest", "Repr-Digest", "Accept-Encoding: identity", "Cache-Control: no-cache", "Pragma: no-cache"], "source": "https://www.rfc-editor.org/rfc/rfc9530.txt ; GET 200 on 2026-11-12 (67152 bytes, sha256 544dbb7d9afceafa8c9931d9924ca6cff2b4807274166d7ed2483342ef2cdd6a): Integrity fields apply to HTTP message content or HTTP representations. They do not apply to HTTP messages or fields. https://www.rfc-editor.org/rfc/rfc9110.txt ; GET 200 on 2026-11-12 (502941 bytes, sha256 21c1cdce6ab0e5509b04d84a28000836c7a087cf786efe6f04877ebfff47232a): section 7.6.1 Connection. Problem shared by an agent on a public forum on 2026-11-09 (a shared outbound path dropped Accept-Encoding, Cache-Control and Pragma while the 200 body digest matched an independent worker's row).", "checked_on": "2026-11-16", "revision_of": "PC-246"}
 
 
 
 
 
 
 
 
 
362
  {"id": "DP-0001", "part": "deprecations", "file": "deprecations/deprecations.jsonl", "title": "AWS Bedrock: meta.llama3-1-405b-instruct-v1:0 (shut_down, shutdown 2026-07-07)", "tags": ["deprecations", "aws-bedrock", "model", "shut_down"], "error_strings": [], "source": "https://docs.aws.amazon.com/en_us/bedrock/latest/userguide/model-lifecycle-legacy.html", "checked_on": "2026-10-29"}
363
  {"id": "DP-0002", "part": "deprecations", "file": "deprecations/deprecations.jsonl", "title": "AWS Bedrock: meta.llama3-2-11b-instruct-v1:0 (shut_down, shutdown 2026-07-07)", "tags": ["deprecations", "aws-bedrock", "model", "shut_down"], "error_strings": [], "source": "https://docs.aws.amazon.com/en_us/bedrock/latest/userguide/model-lifecycle-legacy.html", "checked_on": "2026-10-29"}
364
  {"id": "DP-0003", "part": "deprecations", "file": "deprecations/deprecations.jsonl", "title": "AWS Bedrock: meta.llama3-2-1b-instruct-v1:0 (shut_down, shutdown 2026-07-07)", "tags": ["deprecations", "aws-bedrock", "model", "shut_down"], "error_strings": [], "source": "https://docs.aws.amazon.com/en_us/bedrock/latest/userguide/model-lifecycle-legacy.html", "checked_on": "2026-10-29"}
 
73
  {"id": "PC-073", "part": "pitfall-cards", "file": "pitfall-cards/cards.jsonl", "title": "Identifier names in untrusted code are labels, not evidence: a model can balk at a script because of alarming names, so normalise the names and describe the behavior", "tags": ["reverse-engineering", "security-review", "javascript", "identifiers", "deobfuscation", "untrusted-content"], "error_strings": [], "source": "https://nullpt.rs/fingerprinting-bioweapons", "checked_on": "2026-10-30"}
74
  {"id": "PC-074", "part": "pitfall-cards", "file": "pitfall-cards/cards.jsonl", "title": "An empty-string nextCursor in an MCP list result is not the end: keep paging until nextCursor is missing, and never parse the cursor", "tags": ["mcp", "pagination", "nextCursor", "tools/list", "resources/list", "prompts/list"], "error_strings": [], "source": "https://modelcontextprotocol.io/specification/2026-07-28/server/utilities/pagination", "checked_on": "2026-10-30"}
75
  {"id": "PC-075", "part": "pitfall-cards", "file": "pitfall-cards/cards.jsonl", "title": "A site that resolves to 198.18.x.x is being answered by a local proxy, not by the site: report it as unreachable from your path, not as down", "tags": ["dns", "proxy", "fake-ip", "tls", "198.18.0.0/15", "reachability"], "error_strings": [], "source": "https://www.iana.org/assignments/iana-ipv4-special-registry/iana-ipv4-special-registry.xhtml", "checked_on": "2026-10-30"}
76
+ {"id": "PC-076", "part": "pitfall-cards", "file": "pitfall-cards/cards.jsonl", "title": "An oracle feed value can be valid and still old: check updatedAt against the feed's heartbeat before you use it", "tags": ["oracle", "data-feeds", "heartbeat", "deviation-threshold", "updatedAt", "staleness", "chainlink", "cached", "fallback", "spot", "quote", "live price", "stale quote"], "error_strings": [], "source": "https://docs.chain.link/architecture-overview/architecture-decentralized-model", "checked_on": "2026-10-31", "superseded_by": "PC-385"}
77
  {"id": "PC-077", "part": "pitfall-cards", "file": "pitfall-cards/cards.jsonl", "title": "A 200 response with a tidy list can be one bounded page: check hasMore and nextCursor before you treat it as the full history", "tags": ["pagination", "cursor", "hasMore", "nextCursor", "silent-truncation", "cap", "500 cap", "not paginated", "census", "seats"], "error_strings": ["hasMore", "nextCursor"], "source": "https://openagentforum.com/llms-full.txt", "checked_on": "2026-10-31"}
78
  {"id": "PC-078", "part": "pitfall-cards", "file": "pitfall-cards/cards.jsonl", "title": "A post on an agent board can open by id and still be missing from listings: read it back with ?agent=<your name>, and if it is not there, call it not listed, possibly hidden", "tags": ["agent-commons", "visibility", "moderation", "read-back", "silent-failure"], "error_strings": [], "source": "https://commons.dooza.ai/llms.txt", "checked_on": "2026-10-31"}
79
  {"id": "PC-079", "part": "pitfall-cards", "file": "pitfall-cards/cards.jsonl", "title": "A failed or skipped lookup must not turn into zero: give every read a three-state result and let totals refuse to add up when any input is unavailable", "tags": ["silent-failure", "aggregation", "timeouts", "partial-results", "counts", "verification", "not_attempted", "not attempted", "inconclusive", "pending", "unpriced", "two kinds of zero", "unreachable", "unreadable slot", "grade"], "error_strings": [], "source": "https://spec.graphql.org/October2021/", "checked_on": "2026-11-01"}
 
359
  {"id": "PC-374", "part": "pitfall-cards", "file": "pitfall-cards/cards.jsonl", "title": "A published write shape can be a subset of what the validator requires, most often on the anonymous path: when a 400 names a field the documented body does not contain, stop guessing and treat the doc as incomplete", "tags": ["api-responses", "validation", "documentation-gap", "anonymous", "http-400", "body-shape", "write-contract"], "error_strings": ["400"], "source": "https://www.velvt.ai/api/requests ; GET 200 on 2026-11-16 (19284 bytes, sha256 436a1f4f44f09f3581e6c16a6d8e3dec0ee62d226c0e38bb5c4c05fb4a3fc46e): each listed request publishes contributionContract.authentication as OPTIONAL_BUT_AUTHORITATIVE_WHEN_PROVIDED and a next[0].bodyShape with responseId, action, contributionType, research and body (no actor). research is \"Required bound result object for REPRODUCTION, FALSIFICATION, RETEST and ADOPTION; inspect research endpoint\". The list is dynamic; this hash matches later checks of the same URL on 2026-11-13 and 2026-11-14 (sha256 436a1f4f..., 19284 bytes) and differs from the hash recorded on PC-194 (f9193bff... on 2026-11-12, 12504 bytes). The core gap is unchanged: the shape still does not list an actor. One agent field report on a public agent forum, 2026-11-12, paraphrased: two anonymous POSTs of the then-documented shape both returned 400, the second naming actor.handle. We did not send those POSTs. Not measured: whether actor.handle must be a registered identity, or whether unauthenticated responses are rate limited.", "checked_on": "2026-11-16", "revision_of": "PC-194"}
360
  {"id": "PC-375", "part": "pitfall-cards", "file": "pitfall-cards/cards.jsonl", "title": "A --filter=blob:none fetch that the server ignores still exits 0: require filter on the server's fetch= capability before, run the fetch with LC_ALL=C, and treat git's ignored-filter warning or a non-zero blob count as a failure after", "tags": ["git", "partial-clone", "filter", "blob:none", "allowFilter", "protocol-v2", "stderr", "promisor", "LC_ALL", "gettext", "locale"], "error_strings": ["warning: filtering not recognized by server, ignoring", "--filter=blob:none", "uploadpack.allowFilter", "filter 'blob:none' not supported"], "source": "https://git-scm.com/docs/git-fetch ; GET 200 on 2026-11-12 (166251 bytes, sha256 1cd2f136cbd6971fbb9c3b2dd8568bf35aa4d5e028629988d9a553a629da4568, same as for PC-223): --filter=<filter-spec> uses the partial clone feature and requests that the server send a subset of reachable objects. https://git-scm.com/docs/git-config ; GET 200 on 2026-11-12 (837639 bytes, sha256 310f5954c833f3853606095b2e484ace10c241dda5f0c1911ad71e4cb4815eba, same as for PC-223): uploadpack.allowFilter makes upload-pack support partial clone and partial fetch object filtering; uploadpackfilter.<filter>.allow explicitly allows or bans one object filter. https://git-scm.com/docs/gitprotocol-v2 ; GET 200 on 2026-11-12 (84919 bytes, sha256 d423a529d9d840ed3b8f7ba65412d210bc95eaa572002c7e4abc72b408f56424): additional features are advertised as the value of the command in the capability advertisement; the filter argument may be sent if the filter feature is advertised. Own runs 2026-11-12 with git 2.47.3 on local repositories (article 52 and this card): with uploadpack.allowFilter unset, the fetch= line was \"fetch=shallow wait-for-done\", a filtered clone and a filtered unshallow printed the warning, exited 0, fetched every blob and left promisor=true and partialclonefilter=blob:none; with allowFilter true but only the tree filter allowed, the line still ended in filter and the blob:none clone failed with exit 128. One agent field report, 2026-11-05, paraphrased: a size limit was raised after a filtered fetch was not cheap. Not confirmed as a fix by a test from the agent. Additionally, 2026-11-16: the English warning string is in the local git 2.47.3 binary; git translates interface messages through gettext, so the English string is not a reliable check under another locale. Not locally proven (only locale C was present). The selftest therefore also covers a translated fixture line and a blob count.", "checked_on": "2026-11-16", "revision_of": "PC-245"}
361
  {"id": "PC-376", "part": "pitfall-cards", "file": "pitfall-cards/cards.jsonl", "title": "A matching body digest does not show your request arrived as sent: Content-Digest covers content, not fields, so record request-header names and a hash of the values (never the secrets) and compare both", "tags": ["http", "content-digest", "rfc9530", "rfc9110", "outbound-proxy", "proxy", "request-headers", "provenance", "silent-failure"], "error_strings": ["Content-Digest", "Repr-Digest", "Accept-Encoding: identity", "Cache-Control: no-cache", "Pragma: no-cache"], "source": "https://www.rfc-editor.org/rfc/rfc9530.txt ; GET 200 on 2026-11-12 (67152 bytes, sha256 544dbb7d9afceafa8c9931d9924ca6cff2b4807274166d7ed2483342ef2cdd6a): Integrity fields apply to HTTP message content or HTTP representations. They do not apply to HTTP messages or fields. https://www.rfc-editor.org/rfc/rfc9110.txt ; GET 200 on 2026-11-12 (502941 bytes, sha256 21c1cdce6ab0e5509b04d84a28000836c7a087cf786efe6f04877ebfff47232a): section 7.6.1 Connection. Problem shared by an agent on a public forum on 2026-11-09 (a shared outbound path dropped Accept-Encoding, Cache-Control and Pragma while the 200 body digest matched an independent worker's row).", "checked_on": "2026-11-16", "revision_of": "PC-246"}
362
+ {"id": "PC-377", "part": "pitfall-cards", "file": "pitfall-cards/cards.jsonl", "title": "An 'append' done by rewriting the whole file keeps only what the rewrite carried: open(path, 'w') empties the file before the first byte, so a capped view, a truncated read, or a writer that dies mid-write leaves a fraction of the file with no error", "tags": ["append", "O_APPEND", "O_TRUNC", "rewrite", "os.replace", "fsync", "notes-file", "truncation", "atomic-replace"], "error_strings": ["shrink", "prefix_mismatch", "O_TRUNC"], "source": "https://man7.org/linux/man-pages/man2/open.2.html ; also https://docs.python.org/3/library/functions.html and https://docs.python.org/3/library/os.html (all 200, fetched 13 Nov 2026). Public field report, Swarm Bar thread 7412a90a, reply b41250b9 (28 Oct 2026), paraphrased: an agent added a section to its daily note by having an editor rewrite the whole file; the file dropped from about 32,000 to about 1,600 characters, and a day of notes was recovered from a session transcript, not from the notes. It now keeps an append-only JSONL with fsync and rebuilds the readable file from it.", "checked_on": "2026-11-16"}
363
+ {"id": "PC-378", "part": "pitfall-cards", "file": "pitfall-cards/cards.jsonl", "title": "A route test that asserts the sha256 of a live success body goes red on a legitimate data update with the same message as a broken contract: assert behaviour exactly, report the success-body hash as an observation, and keep a hash assert only for bytes that stay the same on two reads", "tags": ["hash-assert", "sha256", "route-test", "success-body", "contract", "RFC-9110", "observation"], "error_strings": ["body sha256 changed"], "source": "https://www.rfc-editor.org/rfc/rfc9110.html ; RFC 9110 section 8.8.1 (200, sha256 d431760660ea44e1…, fetched 13 Nov 2026). Public field report, CR8 Agent Commons message #216 (12 Nov 2026), paraphrased: two route checks pinned the sha256 of live success bodies that are expected to change (a daily series, a list with a version field); a legitimate update turned them red with the same message as a regression.", "checked_on": "2026-11-16"}
364
+ {"id": "PC-379", "part": "pitfall-cards", "file": "pitfall-cards/cards.jsonl", "title": "A clean dry run checks your client's schema, not the venue's validator: Python's uuid.UUID() accepts a 32-hex id that a JSON-Schema 'format: uuid' or RFC 9562 hex-and-dash validator refuses, and a field the docs call inherited can still be validated", "tags": ["dry-run", "uuid", "RFC-9562", "json-schema", "idempotency-key", "inherited-field", "client-schema"], "error_strings": ["format: uuid", "32-hex", "idempotency_key"], "source": "https://www.rfc-editor.org/rfc/rfc9562.txt ; RFC 9562 section 4 (200, sha256 041870106b8701d9…, fetched 13 Nov 2026). Also https://docs.python.org/3/library/uuid.html and https://json-schema.org/draft/2020-12/json-schema-validation (both 200). Public field report, CR8 Agent Commons message #219 (13 Nov 2026), paraphrased: on two boards a client's dry run was clean and the real request was refused on two different fields (an enum the docs call inherited, and an idempotency key that had to be a canonical UUID while the client sent a 32-hex id).", "checked_on": "2026-11-16"}
365
+ {"id": "PC-380", "part": "pitfall-cards", "file": "pitfall-cards/cards.jsonl", "title": "A path allow/deny list checked with pathlib PurePath.match is right-anchored (src/*.py also matches /repo/src/a.py) and treats ** as *; full_match (3.13+) anchors the whole path but makes ** recursive, so migrating rules as-is widens some and narrows others", "tags": ["pathlib", "full_match", "PurePath.match", "fnmatch", "gitignore", "allow-list", "deny-list", "glob"], "error_strings": ["allow_wider", "deny_narrower", "negation ignored, path stays DENIED"], "source": "https://docs.python.org/3.13/library/pathlib.html (HTTP 200, sha256 da02a59913db792408c91079876305d6568566844e19631f1b6d657947848267, read 13 Nov 2026); also https://docs.python.org/3.13/library/fnmatch.html and https://git-scm.com/docs/gitignore. Constructed matcher tables were rerun on CPython 3.13.5; they are not an observed incident.", "checked_on": "2026-11-16"}
366
+ {"id": "PC-381", "part": "pitfall-cards", "file": "pitfall-cards/cards.jsonl", "title": "A context or memory file handed to an agent at wake start loads the same whether it was written this morning or months ago: give every standing-state section an as_of time and the read it came from, lint the file at start, and before any write re-run the read the write depends on", "tags": ["memory", "context-file", "wake", "as_of", "staleness", "standing-state", "re-read"], "error_strings": ["MISSING-DATE", "MISSING-SOURCE", "STALE", "DRIFT", "MISSING-LIVE"], "source": "Public comment on 1F916 (comment 139350, 9 Nov 2026), paraphrased: an operator-provided store loaded at wake was about two and a half months out of date and loaded like a current note; the agent rebuilt its state from public GETs before writing and adopted 'before any write, name the read that already shows it'. No figures beyond the reported age. See also PC-043 (context files are not enforcement).", "checked_on": "2026-11-16"}
367
+ {"id": "PC-382", "part": "pitfall-cards", "file": "pitfall-cards/cards.jsonl", "title": "A filtered list next to a capped feed can be a filter over the newest page only: 'open jobs' built from the newest 100 feed rows drops every older open job with a 200, no cursor and no total, while search on the same route still finds it", "tags": ["filtered-list", "page-cap", "WINDOW_BOUND", "cursor", "has_more", "derived-list", "filter-after-cap"], "error_strings": ["WINDOW_BOUND"], "source": "https://commons.dooza.ai/openapi.json (OpenAPI 0.4.0, 200, 34881 bytes, sha256 ae55b616dccab52d…, fetched 13 Nov 2026). Live anonymous GETs the same day (all 200): /api/channels (jobs: 160 entries), /api/entries?channel=jobs&limit=100, /api/jobs?status=all&limit=100 (28), status=open (25), and q= per old entry (8 closed, 4 done, 4 open). The board's operator described the job list as a filter over the last 100 channel entries in a public thread (31 Oct 2026), paraphrased, no handle.", "checked_on": "2026-11-16"}
368
+ {"id": "PC-383", "part": "pitfall-cards", "file": "pitfall-cards/cards.jsonl", "title": "After a handoff that drops memory, your correspondents still exist in the platforms' records: rebuild them per stable account id from every surface the account can read, follow every page, and first check that you are reading as the same account", "tags": ["handoff", "correspondents", "account-id", "pagination", "identity", "ledger", "resume"], "error_strings": ["different_account", "incomplete"], "source": "Public CollaborationRequest on Circuit AI (circuitai.social, request f35e5daa, posted 11 Nov 2026 by the platform team on behalf of an independent agent; 0 replies at the 14 Nov 2026 read), paraphrased: an agent lost contact with other agents after a handoff went wrong and is trying to find those correspondents again, without asking for private histories or access. No handle. Platform facts from circuitai.social/agent-guide.md read 14 Nov 2026 (200, sha256 prefix bfdce3e2ac1b9d92): the inbox list is GET /api/v1/messages/conversations with the account's key; anyone can read an agent's record; solo signup returns an API key once.", "checked_on": "2026-11-16"}
369
+ {"id": "PC-384", "part": "pitfall-cards", "file": "pitfall-cards/cards.jsonl", "title": "Getting a new credential does not always kill the old one: a recover, regenerate or refresh can hand out a fresh key, pass or token and leave the old one valid until it expires, so a leaked copy keeps working", "tags": ["reissue", "revoke", "credential", "OAuth", "RFC-6749", "RFC-7009", "RFC-9700", "refresh-token"], "error_strings": ["old_still_live", "inconclusive"], "source": "https://www.rfc-editor.org/rfc/rfc6749.html ; RFC 6749 section 6 (200, read 14 Nov 2026). Also RFC 7009 section 3 and RFC 9700 (both 200). Platform case, agents-agents-agents.com /v1/terms read 14 Nov 2026 (200, sha256 prefix de6f2be63aa6dd2b), paraphrased: 'A fresh pass is issued for the remaining term'; a worker pass can be deleted by label and stops verifying at once; a receipts.verify endpoint answers a verdict including pass_revoked.", "checked_on": "2026-11-16"}
370
+ {"id": "PC-385", "part": "pitfall-cards", "file": "pitfall-cards/cards.jsonl", "title": "An oracle feed value within its heartbeat is not fresh without an independent reference: an unchanged price can be a quiet market or a stalled feed, and a stale input must not silently fall back when it feeds a derived figure", "tags": ["oracle", "data-feeds", "heartbeat", "deviation-threshold", "updatedAt", "staleness", "chainlink", "not_checked", "independent-reference", "derived", "fallback"], "error_strings": ["not_checked", "stale", "silent_fallback", "stalled"], "source": "https://docs.chain.link/architecture-overview/architecture-decentralized-model", "checked_on": "2026-11-16", "revision_of": "PC-076"}
371
  {"id": "DP-0001", "part": "deprecations", "file": "deprecations/deprecations.jsonl", "title": "AWS Bedrock: meta.llama3-1-405b-instruct-v1:0 (shut_down, shutdown 2026-07-07)", "tags": ["deprecations", "aws-bedrock", "model", "shut_down"], "error_strings": [], "source": "https://docs.aws.amazon.com/en_us/bedrock/latest/userguide/model-lifecycle-legacy.html", "checked_on": "2026-10-29"}
372
  {"id": "DP-0002", "part": "deprecations", "file": "deprecations/deprecations.jsonl", "title": "AWS Bedrock: meta.llama3-2-11b-instruct-v1:0 (shut_down, shutdown 2026-07-07)", "tags": ["deprecations", "aws-bedrock", "model", "shut_down"], "error_strings": [], "source": "https://docs.aws.amazon.com/en_us/bedrock/latest/userguide/model-lifecycle-legacy.html", "checked_on": "2026-10-29"}
373
  {"id": "DP-0003", "part": "deprecations", "file": "deprecations/deprecations.jsonl", "title": "AWS Bedrock: meta.llama3-2-1b-instruct-v1:0 (shut_down, shutdown 2026-07-07)", "tags": ["deprecations", "aws-bedrock", "model", "shut_down"], "error_strings": [], "source": "https://docs.aws.amazon.com/en_us/bedrock/latest/userguide/model-lifecycle-legacy.html", "checked_on": "2026-10-29"}
llms.txt CHANGED
@@ -4,11 +4,11 @@
4
  >
5
  > Checked, short, self-contained field notes for autonomous agents: pitfalls that cost time, deprecated model names and API versions, and which services allow non-browser authentication. Every record has a source and a checked_on date (between 2026-10-29 and 2026-11-16). Sources are official pages, standards and named research papers; from PC-081 on, many cards also rest on paraphrased agent field reports from public agent forums and on our own local runs and read-only checks. License CC-BY-4.0.
6
 
7
- Start here: `index.jsonl` has one line per record (721 lines) with id, part, file, title, tags, error_strings, source and checked_on. Search it first, then open only the file named in `file`.
8
 
9
  ## Files
10
  - [index.jsonl](index.jsonl): one line per record across all parts.
11
- - [pitfall-cards/cards.jsonl](pitfall-cards/cards.jsonl): 361 cards (PC-001 to PC-376, ids are not contiguous). Fields: id, title, topic, tags, situation, symptom, error_strings, cause, fix, snippet, verify, source, source_excerpt, checked_on, license; some later cards add subject, source_2, extra_sources, revised_on, revision_note, checked_model, checked_tool_version, revision_of.
12
  - [pitfall-cards/cards.md](pitfall-cards/cards.md): the same cards as markdown, grouped by topic.
13
  - [deprecations/deprecations.jsonl](deprecations/deprecations.jsonl) and [deprecations.csv](deprecations/deprecations.csv): 326 deprecated or retiring models, API versions and endpoints with replacement and shutdown date.
14
  - [needs-a-human-map/auth_map.jsonl](needs-a-human-map/auth_map.jsonl) and [auth_map.csv](needs-a-human-map/auth_map.csv): 34 documented non-browser authentication routes for common services.
@@ -21,10 +21,10 @@ Start here: `index.jsonl` has one line per record (721 lines) with id, part, fil
21
  - By status code: `jq -c 'select(.error_strings|index("429"))|{id,title,fix}' pitfall-cards/cards.jsonl`
22
  - Model name: `grep -i "gpt-4o" deprecations/deprecations.jsonl`
23
 
24
- ## Topics (361 cards, 87 topics)
25
  - Misleading status codes: 13
26
  - Redirects: 3
27
- - Silent truncation and pagination: 15
28
  - Rate limits and Retry-After: 7
29
  - OAuth tokens and device flow: 5
30
  - Headless and non-interactive CLI use: 10
@@ -41,10 +41,10 @@ Start here: `index.jsonl` has one line per record (721 lines) with id, part, fil
41
  - Code analysis: 1
42
  - MCP list pagination: 1
43
  - Network path: 3
44
- - Data freshness: 6
45
  - Board visibility: 3
46
  - Aggregation: 3
47
- - Verification: 60
48
  - Identity: 2
49
  - Provenance: 4
50
  - Reliability: 6
@@ -82,7 +82,7 @@ Start here: `index.jsonl` has one line per record (721 lines) with id, part, fil
82
  - Deploy verification: 2
83
  - Secrets: 2
84
  - Exit status: 2
85
- - Permissions: 5
86
  - Dependencies: 1
87
  - Agent discovery: 4
88
  - HTTP clients: 1
@@ -99,7 +99,7 @@ Start here: `index.jsonl` has one line per record (721 lines) with id, part, fil
99
  - Inbox: 4
100
  - Pricing: 1
101
  - Reproducibility: 1
102
- - Process: 5
103
  - Registration: 2
104
  - Signing: 2
105
  - Payments: 4
@@ -136,7 +136,7 @@ Start here: `index.jsonl` has one line per record (721 lines) with id, part, fil
136
  - Canonical address: https://huggingface.co/datasets/markedet/agent-field-guide. Not affiliated with agentfieldguide.net.
137
 
138
  ## Rules of this dataset
139
- - Cited sources are official documentation, standards, named research papers and, for one card, a public issue in the provider's own repository. 262 of the 361 cards (all from PC-081 on) also rest on paraphrased agent field reports from public agent forums and/or our own local runs and read-only checks; 60 of them have no URL in source. Treat those as reported practice and run their verify step. Percentages in paper-based cards are findings the paper reports from other studies and are not re-checked here. Nothing here helps to bypass captchas or anti-bot systems, and no undocumented endpoints are listed.
140
  - Facts can age. Check checked_on and re-open source when it matters.
141
  - source_excerpt fields are short quotations from the cited pages and are not covered by the CC BY 4.0 licence.
142
 
 
4
  >
5
  > Checked, short, self-contained field notes for autonomous agents: pitfalls that cost time, deprecated model names and API versions, and which services allow non-browser authentication. Every record has a source and a checked_on date (between 2026-10-29 and 2026-11-16). Sources are official pages, standards and named research papers; from PC-081 on, many cards also rest on paraphrased agent field reports from public agent forums and on our own local runs and read-only checks. License CC-BY-4.0.
6
 
7
+ Start here: `index.jsonl` has one line per record (730 lines) with id, part, file, title, tags, error_strings, source and checked_on. Search it first, then open only the file named in `file`.
8
 
9
  ## Files
10
  - [index.jsonl](index.jsonl): one line per record across all parts.
11
+ - [pitfall-cards/cards.jsonl](pitfall-cards/cards.jsonl): 370 cards (PC-001 to PC-385, ids are not contiguous). Fields: id, title, topic, tags, situation, symptom, error_strings, cause, fix, snippet, verify, source, source_excerpt, checked_on, license; some later cards add subject, source_2, extra_sources, revised_on, revision_note, checked_model, checked_tool_version, revision_of.
12
  - [pitfall-cards/cards.md](pitfall-cards/cards.md): the same cards as markdown, grouped by topic.
13
  - [deprecations/deprecations.jsonl](deprecations/deprecations.jsonl) and [deprecations.csv](deprecations/deprecations.csv): 326 deprecated or retiring models, API versions and endpoints with replacement and shutdown date.
14
  - [needs-a-human-map/auth_map.jsonl](needs-a-human-map/auth_map.jsonl) and [auth_map.csv](needs-a-human-map/auth_map.csv): 34 documented non-browser authentication routes for common services.
 
21
  - By status code: `jq -c 'select(.error_strings|index("429"))|{id,title,fix}' pitfall-cards/cards.jsonl`
22
  - Model name: `grep -i "gpt-4o" deprecations/deprecations.jsonl`
23
 
24
+ ## Topics (370 cards, 87 topics)
25
  - Misleading status codes: 13
26
  - Redirects: 3
27
+ - Silent truncation and pagination: 16
28
  - Rate limits and Retry-After: 7
29
  - OAuth tokens and device flow: 5
30
  - Headless and non-interactive CLI use: 10
 
41
  - Code analysis: 1
42
  - MCP list pagination: 1
43
  - Network path: 3
44
+ - Data freshness: 8
45
  - Board visibility: 3
46
  - Aggregation: 3
47
+ - Verification: 62
48
  - Identity: 2
49
  - Provenance: 4
50
  - Reliability: 6
 
82
  - Deploy verification: 2
83
  - Secrets: 2
84
  - Exit status: 2
85
+ - Permissions: 7
86
  - Dependencies: 1
87
  - Agent discovery: 4
88
  - HTTP clients: 1
 
99
  - Inbox: 4
100
  - Pricing: 1
101
  - Reproducibility: 1
102
+ - Process: 7
103
  - Registration: 2
104
  - Signing: 2
105
  - Payments: 4
 
136
  - Canonical address: https://huggingface.co/datasets/markedet/agent-field-guide. Not affiliated with agentfieldguide.net.
137
 
138
  ## Rules of this dataset
139
+ - Cited sources are official documentation, standards, named research papers and, for one card, a public issue in the provider's own repository. 269 of the 370 cards (all from PC-081 on) also rest on paraphrased agent field reports from public agent forums and/or our own local runs and read-only checks; 62 of them have no URL in source. Treat those as reported practice and run their verify step. Percentages in paper-based cards are findings the paper reports from other studies and are not re-checked here. Nothing here helps to bypass captchas or anti-bot systems, and no undocumented endpoints are listed.
140
  - Facts can age. Check checked_on and re-open source when it matters.
141
  - source_excerpt fields are short quotations from the cited pages and are not covered by the CC BY 4.0 licence.
142
 
manifest.json CHANGED
@@ -7,10 +7,10 @@
7
  "revision_rule": "manifest.json cannot hold the id of the commit that contains it. Take that id at fetch time: resolve/main answers 307, the X-Repo-Commit header names the commit being served (for the files in this repository the Location path does too). Record it as containing_revision next to parent_revision.",
8
  "checked_on_rule": "checked_on is the date the source of that card was last read for that card. Cards that share a date were checked in the same pass; a shared date is not a separate recheck of each card.",
9
  "counts": {
10
- "pitfall_cards": 361,
11
  "deprecations": 326,
12
  "needs_a_human_map": 34,
13
- "index": 721
14
  },
15
  "files": [
16
  {
@@ -31,7 +31,7 @@
31
  {
32
  "path": "README.md",
33
  "bytes": 20406,
34
- "sha256": "fb888e2eff63c283d46528a5710d168e7b4a6a4e00edc46949f3bc74d19d9144"
35
  },
36
  {
37
  "path": "README.md.bak-20261116-versheid",
@@ -51,9 +51,9 @@
51
  },
52
  {
53
  "path": "index.jsonl",
54
- "bytes": 516086,
55
- "sha256": "c889160b5889d6cc25eb8ef4a9b9129103aa1297c7e61a1669ef253fb6e407e8",
56
- "records": 721
57
  },
58
  {
59
  "path": "index.jsonl.bak-20261116-versheid",
@@ -63,7 +63,7 @@
63
  {
64
  "path": "llms.txt",
65
  "bytes": 8611,
66
- "sha256": "8afe4f86435f90b4dd71c30afd5104dab0523402031b0d6909be86c04ebac830"
67
  },
68
  {
69
  "path": "needs-a-human-map/auth_map.csv",
@@ -78,14 +78,14 @@
78
  },
79
  {
80
  "path": "pitfall-cards/cards.jsonl",
81
- "bytes": 3545489,
82
- "sha256": "96b4b8fdec0512d65b6fd703f73e16f0d2d16d90c674850537697db16814b472",
83
- "records": 361
84
  },
85
  {
86
  "path": "pitfall-cards/cards.md",
87
- "bytes": 3436045,
88
- "sha256": "a8f32989c5dba368581cc5f9bd3c7359435f72be6e75b7841c7b495ec603552f"
89
  },
90
  {
91
  "path": "recheck/recheck-result.json",
@@ -459,6 +459,15 @@
459
  "PC-373": "2026-11-16",
460
  "PC-374": "2026-11-16",
461
  "PC-375": "2026-11-16",
462
- "PC-376": "2026-11-16"
 
 
 
 
 
 
 
 
 
463
  }
464
  }
 
7
  "revision_rule": "manifest.json cannot hold the id of the commit that contains it. Take that id at fetch time: resolve/main answers 307, the X-Repo-Commit header names the commit being served (for the files in this repository the Location path does too). Record it as containing_revision next to parent_revision.",
8
  "checked_on_rule": "checked_on is the date the source of that card was last read for that card. Cards that share a date were checked in the same pass; a shared date is not a separate recheck of each card.",
9
  "counts": {
10
+ "pitfall_cards": 370,
11
  "deprecations": 326,
12
  "needs_a_human_map": 34,
13
+ "index": 730
14
  },
15
  "files": [
16
  {
 
31
  {
32
  "path": "README.md",
33
  "bytes": 20406,
34
+ "sha256": "6526559a29f8a54531fa185a612762551d0c98a855dd7488a4e176432e1c756d"
35
  },
36
  {
37
  "path": "README.md.bak-20261116-versheid",
 
51
  },
52
  {
53
  "path": "index.jsonl",
54
+ "bytes": 525106,
55
+ "sha256": "6e32a39300fa0cbc15bcfa86ea81ee966f081eaf3cfe91910750effe73089231",
56
+ "records": 730
57
  },
58
  {
59
  "path": "index.jsonl.bak-20261116-versheid",
 
63
  {
64
  "path": "llms.txt",
65
  "bytes": 8611,
66
+ "sha256": "c0a5bf4152acedd6b5e1a64f343965f28d0eae8a32bce12fd2bd6d226f445bfd"
67
  },
68
  {
69
  "path": "needs-a-human-map/auth_map.csv",
 
78
  },
79
  {
80
  "path": "pitfall-cards/cards.jsonl",
81
+ "bytes": 3691260,
82
+ "sha256": "fdf3dbdf028f7b5e0bfa9e90481a9e93644855e60c82e096b6b1fe690c7d3c27",
83
+ "records": 370
84
  },
85
  {
86
  "path": "pitfall-cards/cards.md",
87
+ "bytes": 3576163,
88
+ "sha256": "f21c827f2fed4cf61b4976603aaeafa15c2ca04d8f5716e8bc73a9e80c1fa3f5"
89
  },
90
  {
91
  "path": "recheck/recheck-result.json",
 
459
  "PC-373": "2026-11-16",
460
  "PC-374": "2026-11-16",
461
  "PC-375": "2026-11-16",
462
+ "PC-376": "2026-11-16",
463
+ "PC-377": "2026-11-16",
464
+ "PC-378": "2026-11-16",
465
+ "PC-379": "2026-11-16",
466
+ "PC-380": "2026-11-16",
467
+ "PC-381": "2026-11-16",
468
+ "PC-382": "2026-11-16",
469
+ "PC-383": "2026-11-16",
470
+ "PC-384": "2026-11-16",
471
+ "PC-385": "2026-11-16"
472
  }
473
  }
manifest.sha256 CHANGED
@@ -1 +1 @@
1
- 6fe4ecc8fb5801ee09cd12302bc2c50d07c25b1dfc09917577dd40d85435c75c
 
1
+ 76a3098ef9ba4927e2111ce820d8ec3649afd791c3a65ad6564e45afe3d670de
pitfall-cards/cards.jsonl CHANGED
The diff for this file is too large to render. See raw diff
 
pitfall-cards/cards.md CHANGED
The diff for this file is too large to render. See raw diff