pi-braid 0.43.54 → 0.43.58
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/CHANGELOG.md +48 -0
- package/build-info.json +4 -4
- package/docs/LOCAL_DEV.md +1 -1
- package/package.json +1 -1
- package/skills/release-pipeline/SKILL.md +28 -0
package/CHANGELOG.md
CHANGED
|
@@ -96,6 +96,54 @@ Adding a thirteenth extension moves the counts six source claims pin to the mani
|
|
|
96
96
|
schema's recorded line for the `braid.contrib.*` reader; `check:doc-drift` and `check:schema` demanded
|
|
97
97
|
both repairs in the same commit.
|
|
98
98
|
|
|
99
|
+
## Skills — one design loop, and a gate that refuses a private literal (2026-11-12)
|
|
100
|
+
|
|
101
|
+
Two changes in one release, both about a check that did not exist.
|
|
102
|
+
|
|
103
|
+
**The design loop.** The four design skills now name one loop - measure with `design_audit` at 320, 390,
|
|
104
|
+
768 and 1440, name the identity decisions, build, re-measure, compare - written once in the validator and
|
|
105
|
+
referenced by the creator, the token contract and the landing skill rather than restated in four voices. Two
|
|
106
|
+
properties are new: a design claim without a pasted before/after table does not count, and the loop no longer
|
|
107
|
+
terminates at "no findings" - a finished page must show three identity decisions, which is the check that
|
|
108
|
+
separates a designed page from an undressed template that happens to be clean. Selection is measured rather
|
|
109
|
+
than asserted: a sixteen-prompt fixture scores the shipped catalog before and after (10 of 16 design prompts
|
|
110
|
+
selected nothing before; 14 of 16 select a design skill now, with 0 of 12 guardrail prompts selecting one and
|
|
111
|
+
a control arm that selects none when the four skills are removed), and the labelled relevance set holds at
|
|
112
|
+
recall@2 13/24 with the same 2/7 slot spend.
|
|
113
|
+
|
|
114
|
+
**The private-literal gate.** A leak is refused before it is pushed: the tracked tree, the commits a branch
|
|
115
|
+
adds (subject, body, author and committer addresses) and a tag's annotation, with the pattern list as a
|
|
116
|
+
module built from fragments so the gate does not refuse its own source. An allowance covers a declared line
|
|
117
|
+
and is printed, so it cannot hide a second one.
|
|
118
|
+
|
|
119
|
+
## Release — a published version names a content, and CI now compares it (2026-11-12)
|
|
120
|
+
|
|
121
|
+
The publish loop skips a workspace whose version is already on the registry, so a fix that landed in a
|
|
122
|
+
workspace whose version did not move was published as nothing. Three fixes were in that state and stayed
|
|
123
|
+
unshipped for a day while CI was green.
|
|
124
|
+
|
|
125
|
+
`release-ledger.json` now holds, per `name@version`, a digest of the packed file contents and the commit it
|
|
126
|
+
was released from, and two things refuse without it: `npm run ci:check` (a recorded version whose content
|
|
127
|
+
moved is a refusal that names the changed paths and the remedy) and the publish workflow after it publishes
|
|
128
|
+
(`release:content:verify` compares the registry's own answer against the ledger). A version the ledger does
|
|
129
|
+
not hold is a new version and says so. The digest is per file rather than per tarball, because npm's packing
|
|
130
|
+
is not byte-reproducible here: the same tree packs to the same file contents and a different tarball, and
|
|
131
|
+
the fixture proves both halves of that sentence.
|
|
132
|
+
|
|
133
|
+
## Skills — a recovery table that lists classes, not one command per platform (2026-11-12)
|
|
134
|
+
|
|
135
|
+
The macOS row of the VPN skill's recovery table named the client's URL links and nothing else, while the
|
|
136
|
+
Windows and Linux rows named their operating system's commands. That shape - one command per platform -
|
|
137
|
+
reads as complete: a reader who finds their platform's row stops looking. It cost hours, because the
|
|
138
|
+
installed client carried no connect link while an OS-visible NetworkExtension profile was startable by
|
|
139
|
+
name the whole time.
|
|
140
|
+
|
|
141
|
+
The section now enumerates four classes of control surface (the OS profile, the client's links, the
|
|
142
|
+
client's CLI, process state for diagnosis only) in an order that says why it is that order, and the stop
|
|
143
|
+
condition - hand the machine to its owner - is allowed only with the four-class table filled in: the
|
|
144
|
+
command run and its answer per class. The arm-state roster is labelled for what it is, evidence derived
|
|
145
|
+
from installs on one machine, against a procedure that carries over to a machine nobody has seen.
|
|
146
|
+
|
|
99
147
|
## Skills — the VPN skill names the surface that actually starts a tunnel (2026-11-12)
|
|
100
148
|
|
|
101
149
|
A VPN profile was disconnected for hours and the operator clicked Connect by hand, because the skill's
|
package/build-info.json
CHANGED
|
@@ -1,9 +1,9 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "pi-braid",
|
|
3
|
-
"version": "0.43.
|
|
4
|
-
"sha": "
|
|
5
|
-
"shaShort": "
|
|
3
|
+
"version": "0.43.58",
|
|
4
|
+
"sha": "99f15692c030acf8aaba4f3c42b15321489b74e2",
|
|
5
|
+
"shaShort": "99f15692",
|
|
6
6
|
"branch": "detached",
|
|
7
7
|
"dirty": false,
|
|
8
|
-
"packedAt": "2026-11-
|
|
8
|
+
"packedAt": "2026-11-12T12:05:14.037Z"
|
|
9
9
|
}
|
package/docs/LOCAL_DEV.md
CHANGED
|
@@ -138,7 +138,7 @@ before and after a full run. A non-zero delta is the leak this rule prevents.
|
|
|
138
138
|
### New fixtures: use `node:test`
|
|
139
139
|
|
|
140
140
|
A new fixture is written on `node:test`. The runner does not care which style a
|
|
141
|
-
file uses -
|
|
141
|
+
file uses - 121 of the core suite's 357 fixtures already use `node:test` - so the
|
|
142
142
|
reason is the failure the other style produces. A hand-rolled helper that calls
|
|
143
143
|
`fn()` without awaiting it discards the case's promise, so an `async` case runs
|
|
144
144
|
detached and the file can print its green tally before that case settles, with
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "pi-braid",
|
|
3
|
-
"version": "0.43.
|
|
3
|
+
"version": "0.43.58",
|
|
4
4
|
"description": "🥨 Braid - Metacognitive extension suite for pi coding agent: Adds memory, plan tracking, structured cognitive signals and completion checks on top of pi core loop, so long sessions keep their context and stop claiming unfinished work.",
|
|
5
5
|
"type": "module",
|
|
6
6
|
"main": "./dist/index.js",
|
|
@@ -176,8 +176,36 @@ succeeds and that package stays at its old version on the registry. Bump the
|
|
|
176
176
|
sibling's `package.json` and commit it before dispatching, or the fix it
|
|
177
177
|
carries ships in the repository and nowhere else.
|
|
178
178
|
|
|
179
|
+
That skip is also why a version's content is recorded. `release-ledger.json`
|
|
180
|
+
holds one entry per published `name@version`, listing every file the tarball
|
|
181
|
+
carries with its sha256, and the write loop refuses before the registry read
|
|
182
|
+
when a recorded version's content has moved or a version has no entry at all.
|
|
183
|
+
The record is written at release time, never after publishing:
|
|
184
|
+
|
|
185
|
+
```bash
|
|
186
|
+
npm run release:content:record # after the version bump, before the tag
|
|
187
|
+
npm run release:content:check # what the workflow runs, if you want it early
|
|
188
|
+
```
|
|
189
|
+
|
|
190
|
+
Commit `release-ledger.json` with the bump. It records the tree, so a version
|
|
191
|
+
that is recorded and then edited becomes unpublishable until it is bumped;
|
|
192
|
+
that is the point, since the edit could not reach the registry under the same
|
|
193
|
+
version string. After the workflow publishes, its last comparison is
|
|
194
|
+
`release:content:verify`, which holds each published artifact to the ledger;
|
|
195
|
+
that step covers the versions a run skips, and fails the run when the registry
|
|
196
|
+
serves content the ledger does not hold.
|
|
197
|
+
|
|
198
|
+
A ledger that predates a package's first entry is seeded from the registry,
|
|
199
|
+
which is the only truthful source for versions that already shipped:
|
|
200
|
+
|
|
201
|
+
```bash
|
|
202
|
+
npm run release:content:record -- --seed-from-registry
|
|
203
|
+
```
|
|
204
|
+
|
|
179
205
|
Preparation:
|
|
180
206
|
|
|
207
|
+
- Record each stable version with `npm run release:content:record` and commit
|
|
208
|
+
the ledger alongside it.
|
|
181
209
|
- Commit the intended stable package versions. Refresh `package-lock.json` as
|
|
182
210
|
a release convention, as described in `CONTRIBUTING.md`.
|
|
183
211
|
- Configure trusted publishing for the package on npmjs.com.
|