You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Commit 4b6ff1a
Browse filesBrowse the repository at this point in the historyBrowse files
chore(sep-2640): re-quote nine trimmed rows and four paraphrases against the Final text
Addresses the review on #511 (2026-10-08).
- Nine rows trimmed in the first pass still carried pre-copy-edit wording
(em-dash asides the Final text turned into commas, colons and
parentheses). #513's drift check only catches whole-sentence quotes, so
it could not see them. All nine now quote f56f204f verbatim.
- A sentence-by-sentence check against f56f204f found four more rows that
paraphrased rather than quoted: host-origin-tag-visible,
host-cache-write-isolation, host-cache-excluded-from-discovery and
host-ignore-allowed-tools. The last had also dropped the "unless the user
has explicitly approved that grant" exception, which changed its meaning.
- capability-declaration-inline and host-frontmatter-comparison now quote
the Final sentence instead of a gloss on it.
- skills-get-cache-attributes quotes the ext-skills stable page, because
the SEP leaves that question open. The header now names that second
source, and the row's url points at the stable page.
- The header now counts fd72d3f1 as normative: it added the
resources-capability MUST as new text, not as a restatement of
Dependencies.
mcpkit examples/skills: 45 pass / 0 fail on this branch.
Copy file name to clipboardExpand all lines: src/seps/sep-2640.yaml
+42-24Lines changed: 42 additions & 24 deletions
Original file line number
Diff line number
Diff line change
@@ -26,16 +26,32 @@
26
26
# final pass: that pass is this one. SEP-2640 was marked Accepted on 2026-09-03
27
27
# (d6b31a03) and Final on 2026-09-11 (8075db4a), and the source moved from the
28
28
# sep/skills-extension branch to main. Four commits landed on the text between
29
-
# a3e147ca and f56f204f, only one of them normative: the Skill Format section
29
+
# a3e147ca and f56f204f, two of them normative. The Skill Format section
30
30
# gained a paragraph binding clients to the Agent Skills specification's own
31
-
# backward-compatibility mechanisms, declared below as two excluded rows.
32
-
# The Capability Declaration section was reworded to locate the declaration in
33
-
# the `extensions` field of the server's capabilities rather than loosely in
34
-
# the `initialize` response, and the `resources`-capability requirement is now
35
-
# stated there outright instead of being derived from Dependencies; both rows
36
-
# are restated from the Final text. A copy-editing pass (582d814a) replaced
37
-
# em-dashes with commas, colons and parentheses throughout, which left twelve
38
-
# rows quoting text the SEP no longer contains, so those quotes are re-synced.
31
+
# backward-compatibility mechanisms, declared below as two excluded rows. And
32
+
# fd72d3f1 (2026-09-13) added a new MUST to Capability Declaration: a server
33
+
# declaring this extension MUST also declare the `resources` capability. That
34
+
# sentence did not exist before fd72d3f1; it is new text, not a restatement of
35
+
# Dependencies (sep-2640-capability-requires-resources). The same commit moved
36
+
# the declaration from the `initialize` response to the `extensions` field of
37
+
# the server's capabilities (sep-2640-capability-declaration-inline). A copy-editing pass (582d814a) replaced
38
+
# em-dashes with commas, colons and parentheses throughout, which left rows
39
+
# quoting text the SEP no longer contains. The first pass re-synced twelve. A
40
+
# review on conformance#511 (2026-10-08) found nine more that #513's drift
41
+
# check could not see, because they had been trimmed rather than quoted whole,
42
+
# and a sentence-by-sentence check against f56f204f turned up four rows that
43
+
# paraphrased rather than quoted (origin-tag-visible, ignore-allowed-tools,
44
+
# cache-write-isolation, cache-excluded-from-discovery). All are now verbatim.
45
+
# host-ignore-allowed-tools had also dropped the "unless the user has
46
+
# explicitly approved that grant" exception, which changed its meaning.
47
+
#
48
+
# second source: one row quotes the ext-skills stable page rather than the SEP.
49
+
# The Final SEP leaves caching on `skills/get` open ("whether the result should
50
+
# also carry the base protocol's caching attributes ... is left open").
51
+
# modelcontextprotocol/ext-skills#139 (2026-09-10) settled it on the stable
52
+
# page, specification/stable/skills.mdx, which the WG treats as the source of
53
+
# truth after Final. sep-2640-skills-get-cache-attributes quotes that page and
54
+
# its url points there. Every other row quotes the SEP.
39
55
# The design-history links moved with the repository, from
40
56
# experimental-ext-skills to ext-skills/docs/archive.
41
57
#
@@ -101,7 +117,7 @@ requirements:
101
117
- check: sep-2640-nested-skillmd-not-acted-on
102
118
text: "From the enclosing skill's perspective, a nested skill's directory and files are ordinary supporting files, and reading them is ordinary reading. A nested `SKILL.md` read this way is ordinary markdown: hosts MUST NOT act on its frontmatter."
103
119
- check: sep-2640-nested-publication-flat
104
-
text: "A nested skill is published like any other: through its own `skills/list` entry, or by explicit reference. The listing remains flat — an entry for a nested skill is an ordinary entry whose `uri` happens to share a path prefix with the enclosing skill's, and nothing in the listing marks nesting."
120
+
text: "A nested skill is published like any other: through its own `skills/list` entry, or by explicit reference. The listing remains flat: an entry for a nested skill is an ordinary entry whose `uri` happens to share a path prefix with the enclosing skill's, and nothing in the listing marks nesting."
105
121
106
122
# === Resource Metadata ===
107
123
- check: sep-2640-skillmd-mimetype
@@ -119,7 +135,7 @@ requirements:
119
135
# identifiers to per-extension settings objects" and defines no envelope; the
120
136
# SEP-2640 capability block matches it.
121
137
- check: sep-2640-capability-declaration-inline
122
-
text: 'Per SEP-2133 extension negotiation, servers declare support for this extension in the `extensions` field of their capabilities, mapping the extension identifier directly to its per-extension settings object: `{"capabilities": {"extensions": {"io.modelcontextprotocol/skills": {"directoryRead": true}}}}`.'
138
+
text: 'Per SEP-2133 extension negotiation, servers declare support for this extension in the `extensions` field of their capabilities: `{"capabilities": {"extensions": {"io.modelcontextprotocol/skills": {"directoryRead": true}}}}`.'
text: 'Declaring the extension itself commits the server to `skills/list` and `skills/get`.'
@@ -139,7 +155,7 @@ requirements:
139
155
- check: sep-2640-skills-list-entry-atomic
140
156
text: "An entry is atomic: a skill's `resources` set is never split across pages."
141
157
- check: sep-2640-skills-list-cache-attributes
142
-
text: "In protocol versions 2026-07-28 and later, the result also carries the base protocol's list-caching attributes — `ttlMs` and `cacheScope`, as defined for `tools/list` and `resources/list` (SEP-2549) — with the same semantics: a freshness hint for the listing and a cache-scope marker, not an integrity property."
158
+
text: "In protocol versions 2026-07-28 and later, the result also carries the base protocol's list-caching attributes, `ttlMs` and `cacheScope`, as defined for `tools/list` and `resources/list` (SEP-2549), with the same semantics: a freshness hint for the listing and a cache-scope marker, not an integrity property."
143
159
- check: sep-2640-skills-list-may-be-partial
144
160
text: 'A server whose skill catalog is large, generated on demand, or otherwise unenumerable MAY return an empty or partial listing.'
text: "`skills[].uri` (Yes): Resource URI of the skill's `SKILL.md`."
153
169
- check: sep-2640-entry-frontmatter-required
154
-
text: "`frontmatter` is the skill's `SKILL.md` YAML frontmatter rendered verbatim as a JSON object — every field the author wrote, not a curated subset. Because the Agent Skills specification requires `name` and `description`, those fields are always present."
170
+
text: "`frontmatter` is the skill's `SKILL.md` YAML frontmatter rendered verbatim as a JSON object. It contains every field the author wrote, not a curated subset. Because the Agent Skills specification requires `name` and `description`, those fields are always present."
155
171
- check: sep-2640-entry-frontmatter-identical
156
172
text: 'The `frontmatter` object MUST be identical in content to the frontmatter of the `SKILL.md` it describes.'
text: 'If the URI does not identify a skill the server serves, the server MUST return error `-32602` (Invalid params), the same code `resources/read` uses for unknown resources.'
201
217
- check: sep-2640-skills-get-answers-unlisted
202
218
text: 'A server MUST answer for every skill it serves, whether or not that skill appears in its `skills/list` result. A skill absent from a partial listing is still retrievable by URI.'
219
+
# Quotes the ext-skills stable page, not the SEP (see "second source" above).
text: "`GetSkillResult` extends `CacheableResult`, so `ttlMs` and `cacheScope` are **REQUIRED**, as they are on `resources/read`, and carry the same semantics. `ttlMs` is the server's hint for how long a host may treat the entry as current before re-calling `skills/get`. Neither field is an integrity property."
205
223
- check: sep-2640-skills-get-no-cursor
206
224
text: 'The result carries no pagination cursor: a single entry is not a list.'
@@ -217,24 +235,24 @@ requirements:
217
235
- check: sep-2640-host-no-prefetch
218
236
text: "Hosts MUST NOT retrieve a skill's files ahead of need, whether on connection, on listing, or at approval. A `SKILL.md` is fetched when the skill is loaded, and a supporting file when it is read."
219
237
- check: sep-2640-host-frontmatter-comparison
220
-
text: "After fetching a `SKILL.md` for which the host holds an entry, hosts MUST parse its YAML frontmatter and compare it field-by-field against the entry's `frontmatter`. Any discrepancy MUST be treated as a verification failure equivalent to a digest mismatch, and the skill MUST NOT be loaded."
238
+
text: 'After fetching a `SKILL.md` for which the host holds an entry, from either `skills/list` or `skills/get` (digest-verified when the entry''s `resources` is an array, and unverifiable when it is `"dynamic"`), hosts MUST parse its YAML frontmatter and compare it field-by-field against the entry''s `frontmatter`. Any discrepancy MUST be treated as a verification failure equivalent to a digest mismatch, and the skill MUST NOT be loaded.'
text: 'Digests are unsigned and supplied by the same server that supplies the content. A match proves the two are consistent, not that either is trustworthy. Hosts MUST NOT treat a digest match as a security boundary.'
223
241
- check: sep-2640-host-skill-identity-pair
224
-
text: "The identity of an MCP-served skill is the pair of the host's identity for the originating server and the skill's `uri`. Hosts MUST preserve both halves wherever a skill is recorded or addressed — the registry, persisted approvals, the cache, and any tool or path through which the model reaches the skill — and MUST NOT key any of these on the `uri` alone."
242
+
text: "The identity of an MCP-served skill is therefore the pair of the host's identity for the originating server and the skill's `uri`. Hosts MUST preserve both halves wherever a skill is recorded or addressed, including the registry, persisted approvals, the cache, and any tool or path through which the model reaches the skill, and MUST NOT key any of these on the `uri` alone."
225
243
- check: sep-2640-host-not-skill-by-scheme
226
244
text: 'A host MUST NOT conclude that a resource is a skill merely because its URI carries a particular scheme.'
227
245
- check: sep-2640-host-cross-origin-no-shadow
228
246
text: 'When skills from different origins collide on `name`, hosts MUST resolve the name within a per-origin namespace, identifying servers by a host-assigned label; an MCP-served skill MUST NOT silently shadow, or be silently substituted for, a same-named skill from any other origin.'
229
247
- check: sep-2640-host-content-bound-approval
230
-
text: "When a host persists any per-skill user approval, it MUST be bound to the entry's `resources` set — every `uri` and `digest` — observed at the moment of approval. If a subsequent entry for that skilladvertises a different set, the host MUST treat the prior approval as revoked and re-prompt before loading or executing."
248
+
text: "When a host persists any per-skill user approval, it MUST be bound to the entry's `resources` set (every `uri` and `digest`) observed at the moment of approval. If a subsequent entry for that skill, from `skills/list` or `skills/get`, advertises a different set, whether a file was rotated, added, or removed, the host MUST treat the prior approval as revoked and re-prompt before loading or executing."
231
249
- check: sep-2640-host-dynamic-not-content-bound
232
250
text: 'A skill whose `resources` is `"dynamic"` cannot be content-bound: hosts MAY decline to load it, and MUST NOT treat a persisted approval as covering whatever content the server currently serves.'
233
251
- check: sep-2640-host-nested-fresh-consent
234
-
text: 'Activating a nested skill — loading it as a skill in its own right — requires fresh, explicit user consent; approval of the enclosing skill does not substitute for it. A nested `SKILL.md` read as supporting content is ordinary markdown: hosts MUST NOT act on its frontmatter.'
252
+
text: "Activating a nested skill (loading it as a skill in its own right, whether through the host's skill-loading machinery or by giving effect to its frontmatter) requires fresh, explicit user consent; approval of the enclosing skill does not substitute for it."
235
253
236
254
- check: sep-2640-host-origin-tag-visible
237
-
text: 'Hosts MUST tag MCP-served skill content with its originating server identity at the point it enters model context, and MUST NOT present it as indistinguishable from a local filesystem skill.'
255
+
text: 'Hosts MUST tag MCP-served skill content with its originating server identity at the point it enters model context and MUST NOT present an MCP-served skill to the model as indistinguishable from a local filesystem skill.'
text: 'Hosts MUST treat MCP-served skill content as untrusted model input, subject to the same prompt-injection defenses applied to any server-provided text. A server being connected does not make its skill content authoritative.'
@@ -267,16 +285,16 @@ requirements:
267
285
text: "Hosts MUST NOT honor frontmatter fields that widen the model's tool or filesystem permissions when the skill arrives over MCP."
text: 'The Agent Skills `allowed-tools` field, which a filesystem-sourced skill uses to declare the tools available while it runs, MUST be ignored for MCP-origin skills.'
288
+
text: 'The Agent Skills `allowed-tools` field, which a filesystem-sourced skill uses to declare the tools available while it runs, MUST be ignored for MCP-origin skills unless the user has explicitly approved that grant for that skill.'
text: 'Intermediaries MAY attach provenance or verification annotations via `_meta` under their own reverse-domain prefix — not the `io.modelcontextprotocol.skills/` prefix reserved for this extension.'
291
+
text: 'Intermediaries MAY attach provenance or verification annotations via `_meta` under their own reverse-domain prefix, not the `io.modelcontextprotocol.skills/` prefix reserved for this extension.'
text: 'Hosts that cache skill content on disk MUST do one of the following for every file served from the cache: keep the cache where nothing but the host can write to it, or re-hash the file against the entryon every read.'
294
+
text: "Hosts that cache skill content on disk MUST therefore do one of the following for every file served from the cache: keep the cache where nothing but the host can write to it (not the model, not scripts or tools the model runs, not other users of the machine) and never modify a cached file in place; or recompute the file's SHA-256 digest from the cached bytes on each access and compare it against the entry's digest, treating a mismatch exactly as a mismatch on a fresh read."
text: 'Hosts that cache MCP-served skill content on disk MUST also do so in a location excluded from every filesystem-skill discovery path, and MUST treat content loaded from there as MCP-origin.'
297
+
text: 'Hosts that cache MCP-served skill content on disk MUST also do so in a location excluded from every filesystem-skill discovery path, and MUST treat content loaded from that location as having arrived over MCP for all purposes of the no-implicit-local-execution rule above, including after host restart and after the originating server is disconnected.'
text: "Hosts SHOULD remove a server's cached skill content when the user removes that server."
@@ -290,7 +308,7 @@ requirements:
290
308
- check: sep-2640-host-surface-directory-read
291
309
text: 'When the originating server declares `directoryRead`, the host SHOULD surface this capability to the model.'
292
310
- check: sep-2640-sdk-convenience-wrappers
293
-
text: "SDK maintainers SHOULD provide affordances that wrap the underlying resource operations in skill-specific terms. The SDK handles: reading `SKILL.md` frontmatter to populate resource metadata, serving file content on `resources/read`, and answering `skills/get` — and, where the server's skill set is bounded, `skills/list` — computing entry digests and sizes from the registered files, and warning when a registered skill exceeds the Limits."
311
+
text: "SDK maintainers SHOULD provide affordances that wrap the underlying resource operations in skill-specific terms. The SDK handles: reading `SKILL.md` frontmatter to populate resource metadata, serving file content on `resources/read`, and answering `skills/get` (and, where the server's skill set is bounded, `skills/list`), computing entry digests and sizes from the registered files, and warning when a registered skill exceeds the Limits."
text: 'Hosts MUST NOT treat the directory result as extending the manifest. While acting on the skill under the held entry, the host MUST NOT read a newly listed child — an unlisted file is a verification failure, exactly as a digest mismatch is — and MUST NOT surface it to the model as a file of the skill.'
327
+
text: 'Hosts MUST NOT treat the directory result as extending the manifest. While acting on the skill under the held entry, the host MUST NOT read the newly listed child (an unlisted file is a verification failure, exactly as a digest mismatch is) and MUST NOT surface it to the model as a file of the skill.'
310
328
311
329
# === Excluded: not observable on the MCP wire ===
312
330
# Added by the 2026-09-13 Skill Format pass, after this file's first
0 commit comments