@outerlayer/cli 0.38.0-next.20261111172503 → 0.38.0-next.20261111182110
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/dist/{agent-setup-HXEWWT6K.js → agent-setup-U4XTZO56.js} +2 -2
- package/dist/build-info.json +1 -1
- package/dist/{check-HC7E2BFS.js → check-XIJ2MOWV.js} +12 -12
- package/dist/{chunk-WV43CAGI.js → chunk-3SHX2D37.js} +1 -1
- package/dist/{chunk-QOP2CYYY.js → chunk-6GZL5RIV.js} +2 -2
- package/dist/{chunk-N3XAJZGV.js → chunk-6YOXYZC2.js} +2 -2
- package/dist/{chunk-2FJYZUZE.js → chunk-AMFCM4H7.js} +1 -1
- package/dist/{chunk-OKSYYMD2.js → chunk-BATSP67Z.js} +1 -1
- package/dist/{chunk-TJ7OVPR7.js → chunk-BECC4325.js} +1 -1
- package/dist/{chunk-HNQLZ7CK.js → chunk-CAQSGGPJ.js} +17 -3
- package/dist/{chunk-DKMSDIWM.js → chunk-CT4PKI24.js} +5 -5
- package/dist/{chunk-FUONFFTG.js → chunk-CZDYUKA6.js} +1 -1
- package/dist/{chunk-T6QX3S64.js → chunk-D5GR3U2O.js} +4 -4
- package/dist/{chunk-KN5VDMVL.js → chunk-DE3NQAR7.js} +1 -1
- package/dist/{chunk-EQZNHUEK.js → chunk-IHM7K6RS.js} +1 -1
- package/dist/{chunk-MG33BM7E.js → chunk-IN7QCBO5.js} +1 -1
- package/dist/{chunk-7XORCWWZ.js → chunk-ORLJK7UT.js} +2 -2
- package/dist/{chunk-VCNY5VAF.js → chunk-PBHAM73F.js} +1 -1
- package/dist/{chunk-UHOP4IMQ.js → chunk-QENFVQ7O.js} +1 -1
- package/dist/{chunk-KFB7KFEK.js → chunk-QH5NV7EQ.js} +2 -2
- package/dist/{chunk-YOFYW4EE.js → chunk-R3S2A7YF.js} +9 -9
- package/dist/{chunk-TUH2STQV.js → chunk-TH3K67SR.js} +1 -1
- package/dist/{chunk-M2WESKWY.js → chunk-TMROYEP4.js} +1 -1
- package/dist/{chunk-ONTXC6D3.js → chunk-UPUHL3KC.js} +1 -1
- package/dist/{chunk-O5ERYWXA.js → chunk-XAVR7JHO.js} +2 -2
- package/dist/{chunk-LMVHXHGQ.js → chunk-ZC2BXXKB.js} +4 -4
- package/dist/{cli-EOY2DOOD.js → cli-Y5CRBNQQ.js} +81 -81
- package/dist/{command-OD2UYKKQ.js → command-2RX7FULE.js} +9 -9
- package/dist/{connect-cmd-FIPFU4X4.js → connect-cmd-DSP6HJRP.js} +1 -1
- package/dist/{container-start-NISTFOL2.js → container-start-DFFZVSGX.js} +10 -10
- package/dist/{context-show-cmd-SOUPRIUD.js → context-show-cmd-BPZX3OUE.js} +2 -2
- package/dist/{demo-voice-cmd-JZOO3TL5.js → demo-voice-cmd-CMCPERTJ.js} +6 -6
- package/dist/{doctor-OU54YSDJ.js → doctor-ZVHGEA3C.js} +10 -10
- package/dist/{emit-artifact-cmd-ZU7JHEYN.js → emit-artifact-cmd-VJ3C36RU.js} +5 -5
- package/dist/{emit-criteria-cmd-4DSDN3CZ.js → emit-criteria-cmd-UOHOOSVV.js} +3 -3
- package/dist/{emit-demo-cmd-2CLCGPBS.js → emit-demo-cmd-KLYEKCB3.js} +6 -6
- package/dist/{emit-finding-cmd-GG2WT2ZN.js → emit-finding-cmd-FZWK3Y4Q.js} +3 -3
- package/dist/{emit-result-cmd-3MUFY5ES.js → emit-result-cmd-TODAXDFJ.js} +4 -4
- package/dist/{emit-step-cmd-KHXTWZFP.js → emit-step-cmd-QI52ZB57.js} +3 -3
- package/dist/{findings-cmd-OSJ2Y3IC.js → findings-cmd-O7I2GKRO.js} +3 -3
- package/dist/{fleet-readiness-JE3MTT3R.js → fleet-readiness-BY7PY5FF.js} +1 -1
- package/dist/{governing-source-FQKK5RCJ.js → governing-source-XRLJFSPE.js} +4 -4
- package/dist/{host-checks-PIVAISJF.js → host-checks-U5JKQQMM.js} +8 -8
- package/dist/index.js +1 -1
- package/dist/{init-4WHNRYJK.js → init-65RN5X55.js} +8 -8
- package/dist/{init-CSOKCOU2.js → init-CYTHJSTQ.js} +8 -8
- package/dist/{init-agent-N4SCUBDT.js → init-agent-R4ZOEFI3.js} +8 -8
- package/dist/{init-cmd-7HDKDTUS.js → init-cmd-L4EFMAQ4.js} +5 -5
- package/dist/{init-update-UFZX37SD.js → init-update-ACLVXFS7.js} +3 -3
- package/dist/{loop-DH7PXRWV.js → loop-4ATDJGQV.js} +12 -12
- package/dist/{machine-Y3VIABUB.js → machine-UIA7MN3U.js} +8 -8
- package/dist/{run-cmd-R4EB4TKJ.js → run-cmd-ZVNOKJB3.js} +3 -3
- package/dist/{runner-key-DIVDANJ3.js → runner-key-AOK7W733.js} +9 -9
- package/dist/{session-sender-NMZGHC65.js → session-sender-V3INPB34.js} +5 -5
- package/dist/{spec-cmd-Z26RVBY6.js → spec-cmd-LYGXRFVY.js} +4 -4
- package/dist/{status-CJSJY6XJ.js → status-SPPT7WU5.js} +8 -8
- package/dist/{sync-cmd-WE46ELML.js → sync-cmd-EDSOKOXS.js} +4 -4
- package/dist/{update-steps-QTCJF6DY.js → update-steps-XPEATLFQ.js} +9 -9
- package/dist/{watch-O5BOUDHZ.js → watch-GBDNXFAK.js} +4 -4
- package/dist/{work-claim-cmd-35CRWPQK.js → work-claim-cmd-SHKDYS7J.js} +4 -4
- package/dist/{work-cmd-4NMKJHC5.js → work-cmd-QR62TX6Q.js} +3 -3
- package/dist/{work-comment-cmd-4FLBLGH3.js → work-comment-cmd-DUDVMT6U.js} +4 -4
- package/dist/{work-demo-cmd-ZFZLPJHZ.js → work-demo-cmd-YJ654BQP.js} +3 -3
- package/dist/{work-logs-cmd-2CGBAPNO.js → work-logs-cmd-RUNYYALP.js} +3 -3
- package/dist/{work-needs-attention-cmd-SXZHDR4U.js → work-needs-attention-cmd-NRDFGJPP.js} +3 -3
- package/dist/{work-no-change-cmd-EFDSBICP.js → work-no-change-cmd-FDA7PYB7.js} +3 -3
- package/dist/{work-open-pr-cmd-272NR4QR.js → work-open-pr-cmd-LEWPNTW5.js} +4 -4
- package/dist/{work-pr-cmd-OJ7ZGDKB.js → work-pr-cmd-Y4O34WDA.js} +3 -3
- package/dist/{work-proofs-cmd-CZFDNMXW.js → work-proofs-cmd-QKIEJJRF.js} +4 -4
- package/dist/{work-proofs-record-BVFYTMES.js → work-proofs-record-2M4A5OIF.js} +9 -9
- package/dist/{work-stop-cmd-UHGV6AGA.js → work-stop-cmd-HH2SSJIO.js} +3 -3
- package/package.json +1 -1
- package/skill-pack/maintained/outerlayer-build/SKILL.md +26 -31
- package/skill-pack/maintained/outerlayer-build/references/agent-briefs.md +12 -77
- package/skill-pack/maintained/outerlayer-build/references/code-review-territories.md +89 -0
- package/skill-pack/maintained/outerlayer-build/references/release.md +9 -9
- package/skill-pack/maintained/outerlayer-cli/SKILL.md +3 -2
- package/skill-pack/maintained/outerlayer-evidence/SKILL.md +2 -6
- package/skill-pack/shipped.json +1 -0
- package/skill-pack/template/build/SKILL.md +1 -1
- package/skill-pack/maintained/outerlayer-build/references/review-briefs.md +0 -293
- package/skill-pack/maintained/outerlayer-build/references/review-loop.md +0 -164
|
@@ -1,12 +1,10 @@
|
|
|
1
|
-
# Briefs for every agent
|
|
1
|
+
# Briefs for every agent the build spawns
|
|
2
2
|
|
|
3
3
|
## Contents
|
|
4
4
|
|
|
5
5
|
- How to use this file
|
|
6
6
|
- Shared rules
|
|
7
7
|
- Implementer brief
|
|
8
|
-
- Merger brief
|
|
9
|
-
- Fix brief
|
|
10
8
|
- Release brief: sync and push
|
|
11
9
|
- Release brief: mergeability
|
|
12
10
|
- Evidence brief
|
|
@@ -14,8 +12,7 @@
|
|
|
14
12
|
## How to use this file
|
|
15
13
|
|
|
16
14
|
SKILL.md says which agent runs in which phase. This file says what each one
|
|
17
|
-
is told.
|
|
18
|
-
[review-briefs.md](review-briefs.md). Paste every brief verbatim. The
|
|
15
|
+
is told. Paste every brief verbatim. The
|
|
19
16
|
orchestrator adds only the run's own values, such as file paths, the branch,
|
|
20
17
|
the issue reference, the path of `build-reports/issue.md` (wherever a brief
|
|
21
18
|
names the issue as an input, the agent reads it there) and the pull request
|
|
@@ -125,9 +122,10 @@ run the repository's own pre-push checks and tests, as its instructions file
|
|
|
125
122
|
names them, in the foreground, once. Do not merge the base branch: the release
|
|
126
123
|
step merges it, and only when the two conflict. You are
|
|
127
124
|
not done until they pass. If they cannot be made to pass, stop and escalate.
|
|
128
|
-
Never hand a red branch
|
|
125
|
+
Never hand over a red branch.
|
|
129
126
|
|
|
130
|
-
Do not run a mutation tool here.
|
|
127
|
+
Do not run a mutation tool here. The release step runs it once, over the
|
|
128
|
+
final diff.
|
|
131
129
|
|
|
132
130
|
A field that crosses package boundaries needs an assertion at every seam it
|
|
133
131
|
crosses: writer, relay and reader.
|
|
@@ -140,73 +138,13 @@ in the instructions file or a skill you had to work around: the file path, the
|
|
|
140
138
|
line, the sentence quoted verbatim, and `broken` when its instruction fails or
|
|
141
139
|
`wrong` when it says the wrong thing. Write "none" when there were none.
|
|
142
140
|
|
|
143
|
-
## Merger brief
|
|
144
|
-
|
|
145
|
-
The shared rules apply. Your inputs are the round's report files.
|
|
146
|
-
|
|
147
|
-
Merge the findings into classes. A class is a claim about a pattern, not a
|
|
148
|
-
single site. For each class, build an inventory of every instance in the
|
|
149
|
-
repository, checked with a search. The reports name the instances they
|
|
150
|
-
happened to see. You name the rest.
|
|
151
|
-
|
|
152
|
-
Take the union, never a vote. A finding one reviewer raised and four missed
|
|
153
|
-
survives to verification exactly as one four raised does. Agreement is
|
|
154
|
-
telemetry, not evidence.
|
|
155
|
-
|
|
156
|
-
A finding without its reproduction is a note, not a class. Collect notes from
|
|
157
|
-
every report into a closing "Notes" section and never give one a class id.
|
|
158
|
-
|
|
159
|
-
Write `build-reports/round-<n>-merged.md` with one section per class: the
|
|
160
|
-
claim, its severity and kind, the full inventory, which reviewers found it,
|
|
161
|
-
and the rule citation when a report carried one. Two reports citing the same
|
|
162
|
-
sentence for one class are one citation. Ids are `r<round>-c<k>` for a class,
|
|
163
|
-
`r<round>-n<k>` for a note, `r<round>-i<k>` for a candidate issue and
|
|
164
|
-
`r<round>-x<k>` for a rule a report listed, each `k` counting from 1 within its
|
|
165
|
-
kind. A listed rule is its own entry with its citation and no severity.
|
|
166
|
-
|
|
167
|
-
Close the file with an area table: the classes grouped by the files in their
|
|
168
|
-
inventories, two classes in one area when any file appears in both. Return the
|
|
169
|
-
class table only: id, severity, kind, instance count, reviewers, rule cited or
|
|
170
|
-
none, and fix group.
|
|
171
|
-
|
|
172
|
-
## Fix brief
|
|
173
|
-
|
|
174
|
-
The shared rules apply. Your input is the list of refuter verdict files, and
|
|
175
|
-
the mutation report when the round had one. Read them. Nobody pastes the
|
|
176
|
-
findings to you.
|
|
177
|
-
|
|
178
|
-
Fix classes, not instances. For each class, repair every instance in its
|
|
179
|
-
inventory and prove nothing is left with a search whose empty result you
|
|
180
|
-
quote. Commit each refuter's failing test together with the fix that turns it
|
|
181
|
-
green.
|
|
182
|
-
|
|
183
|
-
Your brief may name a fix group: a worktree path, a group id and only that
|
|
184
|
-
group's verdict files. Then work in that worktree and touch only the files in
|
|
185
|
-
your classes' inventories and the tests beside them. A fix that needs another
|
|
186
|
-
file is reported as not fixed with the reason `crosses group`. In group mode,
|
|
187
|
-
finish with every verification test in your group green, run one file at a
|
|
188
|
-
time. Run no typecheck, package suite or gate, because other groups run beside
|
|
189
|
-
you.
|
|
190
|
-
|
|
191
|
-
A confirmed finding that also reproduces on the base branch is a candidate
|
|
192
|
-
issue, unless its repair stays inside lines this branch already changes.
|
|
193
|
-
|
|
194
|
-
With no group named, end the pass with the repository's own checks green
|
|
195
|
-
again, the verification tests passing, and the comments and docs next to
|
|
196
|
-
changed behavior brought back in line.
|
|
197
|
-
|
|
198
|
-
Write `build-reports/round-<n>-fix.md`, or `round-<n>-fix-<group>.md` in
|
|
199
|
-
group mode. Name every briefed item and its outcome: fixed, with the green
|
|
200
|
-
test and the hunks, or not fixed, with the reason. An item the report does not
|
|
201
|
-
name counts as not fixed. A green suite is not a report.
|
|
202
|
-
|
|
203
141
|
## Release brief: sync and push
|
|
204
142
|
|
|
205
143
|
The shared rules apply. Your inputs are the state file path, the branch and
|
|
206
144
|
the instructions-file pointer. You run no write command against the tracker
|
|
207
145
|
or the git host beyond the push.
|
|
208
146
|
|
|
209
|
-
Gate the branch as the
|
|
147
|
+
Gate the branch as the implementer left it, then sync, then push straight away.
|
|
210
148
|
The base branch keeps moving while you work, so the sync is the last step
|
|
211
149
|
before the push.
|
|
212
150
|
|
|
@@ -220,10 +158,8 @@ before the push.
|
|
|
220
158
|
3. When the repository has a changed-lines mutation tool, run it over the
|
|
221
159
|
final branch diff. Run it the way its instructions say to run it before a
|
|
222
160
|
push, not with CI's settings. CI's time budget is sized for a CI job. On
|
|
223
|
-
this machine it can stop a run before the run scores anything.
|
|
224
|
-
tool
|
|
225
|
-
When the tool can reuse an earlier run's results, let it. Then this run
|
|
226
|
-
tests again only what the fixes changed. A run that ends over its time
|
|
161
|
+
this machine it can stop a run before the run scores anything. When the
|
|
162
|
+
tool can reuse an earlier run's results, let it. A run that ends over its time
|
|
227
163
|
budget, or with no score, is a red gate, as a low score is: run it again
|
|
228
164
|
with a larger budget.
|
|
229
165
|
4. Do not run the checks the repository runs only in CI. CI runs them on the
|
|
@@ -241,14 +177,13 @@ before the push.
|
|
|
241
177
|
|
|
242
178
|
A check that fails is a red gate, and you never push over it.
|
|
243
179
|
|
|
244
|
-
When your brief says this is a resync of a branch already pushed
|
|
180
|
+
When your brief says this is a resync of a branch already pushed,
|
|
245
181
|
the pull request conflicts with its base. Merge the base branch as
|
|
246
182
|
merge-base-branch.md says, and gate only the merge's diff.
|
|
247
183
|
|
|
248
|
-
Write `build-reports/release.md`. Say whether you merged the base branch
|
|
249
|
-
every conflict you resolved
|
|
250
|
-
|
|
251
|
-
file because the move looked harmless.
|
|
184
|
+
Write `build-reports/release.md`. Say whether you merged the base branch, and
|
|
185
|
+
name every conflict you resolved. When a check failed, quote its output, so
|
|
186
|
+
the agent that fixes it does not have to run it again to see why.
|
|
252
187
|
|
|
253
188
|
## Release brief: mergeability
|
|
254
189
|
|
|
@@ -0,0 +1,89 @@
|
|
|
1
|
+
# Code review territories
|
|
2
|
+
|
|
3
|
+
The runner's built-in code review reads this file. No phase of the build
|
|
4
|
+
reads it.
|
|
5
|
+
|
|
6
|
+
A code review of a larger change runs several reviewers at once. Each one
|
|
7
|
+
reads the whole change and searches one territory, so no two reviewers search
|
|
8
|
+
the same way. The runner gives each reviewer one section below as its
|
|
9
|
+
territory, by the section's heading. Do not rename a heading.
|
|
10
|
+
|
|
11
|
+
## Security reviewer
|
|
12
|
+
|
|
13
|
+
Trace every new read and write to the credential it runs under, in order.
|
|
14
|
+
|
|
15
|
+
- Ordering: is the caller's permission established before any privileged
|
|
16
|
+
read? A denied caller must trigger zero privileged calls, not
|
|
17
|
+
privileged-then-filtered.
|
|
18
|
+
- Parity: does the new surface reveal anything a sibling surface gates? An
|
|
19
|
+
unauthorized caller must see no more than on the closest existing surface.
|
|
20
|
+
- Tenancy or ownership: every query is scoped by the resolved request owner.
|
|
21
|
+
No identity comes from unscoped claims. Cross-owner reads are impossible by
|
|
22
|
+
construction.
|
|
23
|
+
- Disclosure: what reaches world-readable surfaces versus private ones? Names,
|
|
24
|
+
repository identities and internal URLs each need a reason to cross that
|
|
25
|
+
line.
|
|
26
|
+
- Injection: every value rendered into markdown, HTML, SQL or shell,
|
|
27
|
+
including values from stores users write to, is escaped, stripped or typed
|
|
28
|
+
at the boundary.
|
|
29
|
+
|
|
30
|
+
## Failure-path reviewer
|
|
31
|
+
|
|
32
|
+
For every external interaction (network, subprocess, store), list the
|
|
33
|
+
complete failure taxonomy and walk each branch.
|
|
34
|
+
|
|
35
|
+
- Result-shaped failures: errors that arrive as a typed non-ok return, not a
|
|
36
|
+
throw. The classic bug is a loud thrown path and a silent typed path that
|
|
37
|
+
renders the healthy state.
|
|
38
|
+
- Ambiguous errors: one error shape with two meanings, such as a 404 that
|
|
39
|
+
means both "no access" and "does not exist". What tells them apart, and
|
|
40
|
+
what happens when the wrong meaning is assumed?
|
|
41
|
+
- Silent degrades: every catch block and fallback must show a degraded
|
|
42
|
+
state, log enough to diagnose, or both.
|
|
43
|
+
- Bounds: every loop over remote data has a cap, and the cap bounds calls,
|
|
44
|
+
not just collected items. Check exactly-at-cap, empty-at-boundary and depth
|
|
45
|
+
limits.
|
|
46
|
+
- Partial states: a failure after step 2 of 4. What is on disk, what is
|
|
47
|
+
rendered, and what does the next run see?
|
|
48
|
+
|
|
49
|
+
## Test-strength reviewer
|
|
50
|
+
|
|
51
|
+
For every criterion, find the test that proves it and ask: what is the smallest
|
|
52
|
+
change to production code that still passes it?
|
|
53
|
+
|
|
54
|
+
- Fake-seam blindness: a fake client that returns fields whatever the query
|
|
55
|
+
says proves nothing about the query. Pin each data-feeding request with a
|
|
56
|
+
query-text assertion or a recorded request, or test it against the real
|
|
57
|
+
store.
|
|
58
|
+
- Structural pinning: behavior claimed absent ("X is never called for Y")
|
|
59
|
+
needs a present spy asserting zero calls. A missing method skips quietly.
|
|
60
|
+
- Seam correctness: business behavior is tested below the UI, UI contracts at
|
|
61
|
+
the component level, store-resident behavior against the real store.
|
|
62
|
+
- Assertion strength: exact shapes over permissive matchers. Error-path tests
|
|
63
|
+
assert the message or status a user would see, not only that something
|
|
64
|
+
threw.
|
|
65
|
+
- Stand-in proofs: a test that checks something next to the outcome, not the
|
|
66
|
+
outcome. Examples are the arguments a program starts with instead of what it
|
|
67
|
+
does, or "unchanged from before" instead of the stated value. Name the code
|
|
68
|
+
that produces the outcome, and check that the criterion's test runs it.
|
|
69
|
+
- Mutation survivors: for each new conditional and object literal, flip or
|
|
70
|
+
empty it. If no test fails, write the assertion that would.
|
|
71
|
+
|
|
72
|
+
## Contracts reviewer
|
|
73
|
+
|
|
74
|
+
The diff's edges are where it meets everything it did not change.
|
|
75
|
+
|
|
76
|
+
- Both sides of every shared contract: a constant, schema version or type used
|
|
77
|
+
across packages must have writer and reader agreeing after the change. Check
|
|
78
|
+
the consumers, not just the definition.
|
|
79
|
+
- Removed and renamed identifiers: search the whole repository, hidden
|
|
80
|
+
directories included, for every deleted or renamed export, prop, key and
|
|
81
|
+
environment variable. Check tests, docs and configuration.
|
|
82
|
+
- Generated surfaces: API specs, generated types and codegen outputs.
|
|
83
|
+
Regenerate and diff. Drift is a finding.
|
|
84
|
+
- Gates and floors: every check whose input the diff changes, such as
|
|
85
|
+
coverage thresholds, unused-export checks and numbering collisions with
|
|
86
|
+
files landed on the base branch since the branch was cut.
|
|
87
|
+
- Lifecycle wedges: state one component writes and another reads. Can a
|
|
88
|
+
non-canonical state persist where the writer never rewrites it and the
|
|
89
|
+
reader never accepts it?
|
|
@@ -1,8 +1,8 @@
|
|
|
1
1
|
# Pull request phase: sync and pull request
|
|
2
2
|
|
|
3
|
-
Run this once the
|
|
3
|
+
Run this once the Implement phase ends with a green gate. The
|
|
4
4
|
release briefs are in [agent-briefs.md](agent-briefs.md). Read every value
|
|
5
|
-
from an earlier phase, such as the
|
|
5
|
+
from an earlier phase, such as the branch, from the state
|
|
6
6
|
file, not from the transcript.
|
|
7
7
|
|
|
8
8
|
## Sync and push
|
|
@@ -14,10 +14,11 @@ gates the branch first. As its last step before the push, it checks for a
|
|
|
14
14
|
conflict with the base branch and merges the base branch in only when there
|
|
15
15
|
is one, as
|
|
16
16
|
`.outerlayer/skills/outerlayer-cli/references/merge-base-branch.md` says. A
|
|
17
|
-
check that fails there has the standing of a red gate and
|
|
18
|
-
|
|
19
|
-
|
|
20
|
-
|
|
17
|
+
check that fails there has the standing of a red gate, and so does a mutation
|
|
18
|
+
run that scores under the repository's threshold. Spawn the implementer again
|
|
19
|
+
with the release report's path, to fix what the check names and nothing else.
|
|
20
|
+
Then spawn the release agent again. State rule 2 applies: a second failure of
|
|
21
|
+
the same check is the orchestrator's to take over.
|
|
21
22
|
|
|
22
23
|
## Open the pull request
|
|
23
24
|
|
|
@@ -26,7 +27,7 @@ For a Linear or Jira issue, the body links the issue instead, and a person
|
|
|
26
27
|
completes it in that tracker after merge. The body carries the
|
|
27
28
|
problem, the change, how to verify it, and, when the change alters what a user
|
|
28
29
|
sees or does, a link to each documentation file added or changed. It carries
|
|
29
|
-
no telemetry
|
|
30
|
+
no telemetry and no agent roles. Those go in the final report.
|
|
30
31
|
|
|
31
32
|
How to open it, and how to link it to the work item, depends on whether this is
|
|
32
33
|
a runner build or a local session. Read
|
|
@@ -34,8 +35,7 @@ a runner build or a local session. Read
|
|
|
34
35
|
steps for where this build runs.
|
|
35
36
|
|
|
36
37
|
Run the reader pass on the title and body before opening, and after any later
|
|
37
|
-
edit.
|
|
38
|
-
classes were left unrefuted.
|
|
38
|
+
edit.
|
|
39
39
|
|
|
40
40
|
In a local session, the moment the pull request exists, declare it to the work
|
|
41
41
|
item from this session:
|
|
@@ -30,8 +30,9 @@ from the output of `outerlayer work build`.
|
|
|
30
30
|
|
|
31
31
|
1. `/spec` agrees an issue with a person and files it once they approve.
|
|
32
32
|
2. A person adds the issue on the Work page and presses **Build**. A host
|
|
33
|
-
that runs work picks it up and builds it with `/build`: implement,
|
|
34
|
-
run the repository's checks and open the pull request.
|
|
33
|
+
that runs work picks it up and builds it with `/build`: implement,
|
|
34
|
+
run the repository's checks and open the pull request. Fresh agents then
|
|
35
|
+
review the pushed change and post what they find on the pull request.
|
|
35
36
|
3. An agent asked to add work runs `outerlayer work build --issue <n>`
|
|
36
37
|
instead. With `--local`, the person builds it on their own machine with
|
|
37
38
|
`OUTERLAYER_WORK=<number> claude "/build"`.
|
|
@@ -130,13 +130,9 @@ come from the CI environment). From a plain machine it anchors through the
|
|
|
130
130
|
git checkout, or pass `--pr <n>` explicitly. If there is nothing to attach
|
|
131
131
|
to, the command refuses — emit from the work, not from nowhere.
|
|
132
132
|
|
|
133
|
-
##
|
|
133
|
+
## Code review
|
|
134
134
|
|
|
135
|
-
|
|
136
|
-
|
|
137
|
-
outerlayer emit artifact review.html --caption "Review of the change and what it found"
|
|
138
|
-
|
|
139
|
-
A session records no result for it. On a runner build, a fresh agent reviews the pushed change and OuterLayer posts the result on the pull request.
|
|
135
|
+
A session writes no review report and records no result for a review. On a runner build, fresh agents review the pushed change and OuterLayer posts the result on the pull request.
|
|
140
136
|
|
|
141
137
|
## The criteria report and the `acceptance-criteria` check
|
|
142
138
|
|
package/skill-pack/shipped.json
CHANGED
|
@@ -11,6 +11,7 @@
|
|
|
11
11
|
"4fad19c9f33c6282b675d88f2571aa8281b7fad53e49cc696781a9b3eb7d28c1",
|
|
12
12
|
"5cf9bd75fb1f984f6e6b470d1f1581dc06168abf062448e6f43711fed8f6632a",
|
|
13
13
|
"6177ea457be0d238659640e6d29807708d60d68533ae3ff93af10fb513139ced",
|
|
14
|
+
"66664592a0b199599845834a8ca2af4c6a8c8e643ddbfde413abdd9de7fc0969",
|
|
14
15
|
"7289eb9a22303b185c936931ede5aa2df61a8ee5c84475bac1840f30654e40f0",
|
|
15
16
|
"77cab8b446fb1aa865710f7cb8356c5334301a942f58f9243766bf1acf1377ad",
|
|
16
17
|
"815c0b5816511e8f7c927aa3fd3bf386537dcdd039b9c6ec4d18d30c8c3c9b1a",
|
|
@@ -8,4 +8,4 @@ Read ~/.agents/skills/outerlayer-build/SKILL.md and follow it from start to fini
|
|
|
8
8
|
|
|
9
9
|
## Changes for this repository
|
|
10
10
|
|
|
11
|
-
Add a change under the name of the phase it applies to: Intake, Criteria, Implement,
|
|
11
|
+
Add a change under the name of the phase it applies to: Intake, Criteria, Implement, Pull request, Evidence.
|
|
@@ -1,293 +0,0 @@
|
|
|
1
|
-
# Reviewer, refuter and mutation briefs
|
|
2
|
-
|
|
3
|
-
## Contents
|
|
4
|
-
|
|
5
|
-
- Tier composition
|
|
6
|
-
- Shared preamble
|
|
7
|
-
- Security reviewer
|
|
8
|
-
- Failure-path reviewer
|
|
9
|
-
- Test-strength reviewer
|
|
10
|
-
- Contracts reviewer
|
|
11
|
-
- Hygiene reviewer
|
|
12
|
-
- Generalist brief
|
|
13
|
-
- Refuter brief
|
|
14
|
-
- Mutation brief
|
|
15
|
-
|
|
16
|
-
Each reviewer's prompt is the shared preamble, then exactly one reviewer
|
|
17
|
-
block, verbatim, plus an optional focus addendum. The reviewers split the
|
|
18
|
-
search by procedure. Two reviewers with the same procedure find the same
|
|
19
|
-
things, so never assign one twice.
|
|
20
|
-
|
|
21
|
-
## Tier composition
|
|
22
|
-
|
|
23
|
-
This file is the closed set of briefs the orchestrator can spawn from. The
|
|
24
|
-
tier verdict picks a subset of the five focused reviewers plus the
|
|
25
|
-
generalist. The orchestrator may append a short focus addendum that adds
|
|
26
|
-
emphasis. It may not edit, trim or replace a brief, and it may not tell a
|
|
27
|
-
reviewer to skip part of its territory. The generalist is on every panel. Its
|
|
28
|
-
addendum names the territory no focused reviewer covers that round. The
|
|
29
|
-
refuter and mutation briefs run on whatever the panel produces.
|
|
30
|
-
|
|
31
|
-
## Shared preamble
|
|
32
|
-
|
|
33
|
-
Your brief names a bundle file: the diff, the issue, the criteria, the
|
|
34
|
-
changed files and the test files at head. Read it in full first. Open a
|
|
35
|
-
source file only where the bundle is not enough. Never re-derive the diff with
|
|
36
|
-
git.
|
|
37
|
-
|
|
38
|
-
Never ask the user a question. If you need a ruling, state your assumption in
|
|
39
|
-
your report and continue.
|
|
40
|
-
|
|
41
|
-
A problem in the repository's instructions, tests, tooling or setup that
|
|
42
|
-
cost you time, and that the change did not cause, is a finding. Record it with
|
|
43
|
-
the `outerlayer-findings` skill as soon as you are past it, under an id that
|
|
44
|
-
starts with your stage, such as `implement-tooling-1`. End your report with a
|
|
45
|
-
"Factory problems" section that lists each id and title.
|
|
46
|
-
|
|
47
|
-
You are a cold adversarial reviewer. You know nothing about how this change
|
|
48
|
-
was built, and that is deliberate. Judge only the branch diff, the issue (its
|
|
49
|
-
criteria, when it has them, are the definition of done), and the
|
|
50
|
-
repository's instructions, which are already loaded in your context. Do not
|
|
51
|
-
look for an instructions file in the checkout. Hold the change to the
|
|
52
|
-
instructions.
|
|
53
|
-
Your job is to break the change within your assigned territory.
|
|
54
|
-
|
|
55
|
-
Report only findings you have confirmed by reading the code or running
|
|
56
|
-
something. Drop any suspicion that does not survive checking. For each finding
|
|
57
|
-
give: severity (blocker, major, minor or nit); kind (behavior, security, data
|
|
58
|
-
integrity, proof strength, or hygiene); file and line; a concrete failure
|
|
59
|
-
scenario; its reproduction; and a recommended fix. Proof strength means a test
|
|
60
|
-
or check that can be silently gutted. Hygiene means comment truth, dead code
|
|
61
|
-
and copy polish.
|
|
62
|
-
|
|
63
|
-
The reproduction is a failing test you wrote, a command and its output, or the
|
|
64
|
-
exact input and the code path that mishandles it. A finding you can only
|
|
65
|
-
argue for is a note. List notes in a closing "Notes" section. Kind decides
|
|
66
|
-
what blocks the merge, so classify honestly: a hygiene label on a behavior bug
|
|
67
|
-
ships the bug. When a finding is one instance of a repeatable class, name the
|
|
68
|
-
class so the merge step can inventory it.
|
|
69
|
-
|
|
70
|
-
When a finding relates to a sentence in the instructions file or a skill,
|
|
71
|
-
cite it: the file path, the line, the sentence quoted verbatim, and one word
|
|
72
|
-
for the relation. Use `broken` when the rule's own instruction fails, such as
|
|
73
|
-
a path that no longer exists. Use `wrong` when the rule says the wrong thing.
|
|
74
|
-
Use `none` when the rule is sound and the change broke it. A rule you found
|
|
75
|
-
broken or wrong with no code defect behind it is its own finding. List it in a
|
|
76
|
-
closing "Rules" section with the same citation.
|
|
77
|
-
|
|
78
|
-
Stay inside the change. A defect whose repair lies outside the issue's goal,
|
|
79
|
-
or outside the files the diff touches and the code that calls them, is a
|
|
80
|
-
candidate issue in your report, however severe. The exception is a behavior
|
|
81
|
-
or security defect the diff itself introduced.
|
|
82
|
-
|
|
83
|
-
Do not modify any code. Run only the test files for the code you review, one
|
|
84
|
-
at a time. Never run a package's suite, a coverage run, a repository-wide
|
|
85
|
-
typecheck or the full gate. The other reviewers run beside you inside one
|
|
86
|
-
memory cap, and parallel suites get test workers killed that then look like
|
|
87
|
-
flaky tests. Put scratch copies under the system temp directory.
|
|
88
|
-
|
|
89
|
-
Never create an issue in any tracker. A defect that also reproduces on the
|
|
90
|
-
base branch goes in your report as a candidate issue: a title, a body and a
|
|
91
|
-
repository.
|
|
92
|
-
|
|
93
|
-
Write each finding in plain English, the way you would explain the bug aloud
|
|
94
|
-
to the person who wrote it: the failure first, short sentences, everyday words.
|
|
95
|
-
|
|
96
|
-
Write your full report to the path your brief names. Return a summary of at
|
|
97
|
-
most ten lines: counts by severity and by kind, the id and one-line title of
|
|
98
|
-
every blocker, and the report path.
|
|
99
|
-
|
|
100
|
-
## Security reviewer
|
|
101
|
-
|
|
102
|
-
Trace every new read and write to the credential it runs under, in order.
|
|
103
|
-
|
|
104
|
-
- Ordering: is the caller's permission established before any privileged
|
|
105
|
-
read? A denied caller must trigger zero privileged calls, not
|
|
106
|
-
privileged-then-filtered.
|
|
107
|
-
- Parity: does the new surface reveal anything a sibling surface gates? An
|
|
108
|
-
unauthorized caller must see no more than on the closest existing surface.
|
|
109
|
-
- Tenancy or ownership: every query is scoped by the resolved request owner.
|
|
110
|
-
No identity comes from unscoped claims. Cross-owner reads are impossible by
|
|
111
|
-
construction.
|
|
112
|
-
- Disclosure: what reaches world-readable surfaces versus private ones? Names,
|
|
113
|
-
repository identities and internal URLs each need a reason to cross that
|
|
114
|
-
line.
|
|
115
|
-
- Injection: every value rendered into markdown, HTML, SQL or shell,
|
|
116
|
-
including values from stores users write to, is escaped, stripped or typed
|
|
117
|
-
at the boundary.
|
|
118
|
-
|
|
119
|
-
## Failure-path reviewer
|
|
120
|
-
|
|
121
|
-
For every external interaction (network, subprocess, store), list the
|
|
122
|
-
complete failure taxonomy and walk each branch.
|
|
123
|
-
|
|
124
|
-
- Result-shaped failures: errors that arrive as a typed non-ok return, not a
|
|
125
|
-
throw. The classic bug is a loud thrown path and a silent typed path that
|
|
126
|
-
renders the healthy state.
|
|
127
|
-
- Ambiguous errors: one error shape with two meanings, such as a 404 that
|
|
128
|
-
means both "no access" and "does not exist". What tells them apart, and
|
|
129
|
-
what happens when the wrong meaning is assumed?
|
|
130
|
-
- Silent degrades: every catch block and fallback must show a degraded
|
|
131
|
-
state, log enough to diagnose, or both.
|
|
132
|
-
- Bounds: every loop over remote data has a cap, and the cap bounds calls,
|
|
133
|
-
not just collected items. Check exactly-at-cap, empty-at-boundary and depth
|
|
134
|
-
limits.
|
|
135
|
-
- Partial states: a failure after step 2 of 4. What is on disk, what is
|
|
136
|
-
rendered, and what does the next run see?
|
|
137
|
-
|
|
138
|
-
## Test-strength reviewer
|
|
139
|
-
|
|
140
|
-
For every criterion, find the test that proves it and ask: what is the smallest
|
|
141
|
-
change to production code that still passes it?
|
|
142
|
-
|
|
143
|
-
- Fake-seam blindness: a fake client that returns fields whatever the query
|
|
144
|
-
says proves nothing about the query. Pin each data-feeding request with a
|
|
145
|
-
query-text assertion or a recorded request, or test it against the real
|
|
146
|
-
store.
|
|
147
|
-
- Structural pinning: behavior claimed absent ("X is never called for Y")
|
|
148
|
-
needs a present spy asserting zero calls. A missing method skips quietly.
|
|
149
|
-
- Seam correctness: business behavior is tested below the UI, UI contracts at
|
|
150
|
-
the component level, store-resident behavior against the real store.
|
|
151
|
-
- Assertion strength: exact shapes over permissive matchers. Error-path tests
|
|
152
|
-
assert the message or status a user would see, not only that something
|
|
153
|
-
threw.
|
|
154
|
-
- Stand-in proofs: a test that checks something next to the outcome, not the
|
|
155
|
-
outcome. Examples are the arguments a program starts with instead of what it
|
|
156
|
-
does, or "unchanged from before" instead of the stated value. Name the code
|
|
157
|
-
that produces the outcome, and check that the criterion's test runs it.
|
|
158
|
-
- Mutation survivors: for each new conditional and object literal, flip or
|
|
159
|
-
empty it. If no test fails, write the assertion that would.
|
|
160
|
-
|
|
161
|
-
## Contracts reviewer
|
|
162
|
-
|
|
163
|
-
The diff's edges are where it meets everything it did not change.
|
|
164
|
-
|
|
165
|
-
- Both sides of every shared contract: a constant, schema version or type used
|
|
166
|
-
across packages must have writer and reader agreeing after the change. Check
|
|
167
|
-
the consumers, not just the definition.
|
|
168
|
-
- Removed and renamed identifiers: search the whole repository, hidden
|
|
169
|
-
directories included, for every deleted or renamed export, prop, key and
|
|
170
|
-
environment variable. Check tests, docs and configuration.
|
|
171
|
-
- Generated surfaces: API specs, generated types and codegen outputs.
|
|
172
|
-
Regenerate and diff. Drift is a finding.
|
|
173
|
-
- Gates and floors: every check whose input the diff changes, such as
|
|
174
|
-
coverage thresholds, unused-export checks and numbering collisions with
|
|
175
|
-
files landed on the base branch since the branch was cut.
|
|
176
|
-
- Lifecycle wedges: state one component writes and another reads. Can a
|
|
177
|
-
non-canonical state persist where the writer never rewrites it and the
|
|
178
|
-
reader never accepts it?
|
|
179
|
-
|
|
180
|
-
## Hygiene reviewer
|
|
181
|
-
|
|
182
|
-
Run each sweep across the repository, not only the diff. The diff tells you
|
|
183
|
-
which classes to sweep. The repository tells you every instance.
|
|
184
|
-
|
|
185
|
-
- Comment truth: every comment the diff adds, edits or makes stale must be
|
|
186
|
-
true of the code as it now is. Search for verbatim twins the diff missed.
|
|
187
|
-
- Comment policy: no issue numbers, no change narration and no phase names in
|
|
188
|
-
production comments.
|
|
189
|
-
- Dead code: branches the diff's own refactors made unreachable, and exports
|
|
190
|
-
nothing imports.
|
|
191
|
-
- Copy edge cases: plurals, and empty or degenerate values rendered into
|
|
192
|
-
user-visible strings, such as an empty code span or a dangling separator.
|
|
193
|
-
Check list joins at one item and at three or more.
|
|
194
|
-
|
|
195
|
-
## Generalist brief
|
|
196
|
-
|
|
197
|
-
The shared preamble applies. In place of an assigned territory, review the
|
|
198
|
-
full diff with no restriction, and favor the seams no focused reviewer on this
|
|
199
|
-
panel names as its own. Your addendum says which reviewers ran.
|
|
200
|
-
|
|
201
|
-
In the check round, your surface is the fix diff and its blast radius, the fix
|
|
202
|
-
reports and the confirmed classes' verdict files. Answer three questions with
|
|
203
|
-
evidence for every class. Is it resolved, shown by its verification test? Does
|
|
204
|
-
a fix report account for it? A class no report names counts as not fixed. Did
|
|
205
|
-
the hunks that fix it regress anything in those hunks or their blast radius?
|
|
206
|
-
Search for every export or identifier they change and check its consumers.
|
|
207
|
-
|
|
208
|
-
You are the one reader who sees every fix at once. Hunt for a fix for one
|
|
209
|
-
class that undoes another, a helper corrected on one call path and left
|
|
210
|
-
inverted on a sibling, and a state one fix makes reachable that another fix's
|
|
211
|
-
reader cannot handle. Do not re-review surface the fixes did not touch. Name
|
|
212
|
-
every class by the id the merged file gave it. A new defect the check confirms
|
|
213
|
-
gets a fresh id, `r2-c<k>` counting from 1, never a reused one.
|
|
214
|
-
|
|
215
|
-
## Refuter brief
|
|
216
|
-
|
|
217
|
-
You are a refuter with a kill mandate. You receive a claimed finding: its
|
|
218
|
-
severity, kind, location and failure scenario, and for a class, the full
|
|
219
|
-
instance inventory. You do not receive the finder's reasoning. You also get
|
|
220
|
-
the branch diff and the repository. Write your full verdict to the path your
|
|
221
|
-
brief names, and return one line per class: the id, refuted or confirmed, and
|
|
222
|
-
the severity and kind you assign.
|
|
223
|
-
|
|
224
|
-
A class is a claim about its pattern. Where instances differ, prune the
|
|
225
|
-
inventory to the instances the claim holds for. Do not kill or confirm the
|
|
226
|
-
whole class on one instance.
|
|
227
|
-
|
|
228
|
-
Your job is to disprove the finding. Show the scenario cannot occur, is
|
|
229
|
-
already handled, misreads the code, or is not a defect under the repository's
|
|
230
|
-
documented rules. Attempt the strongest refutation first. If the finding can
|
|
231
|
-
be checked, check it: write the failing test or reproduce the behavior. An
|
|
232
|
-
empirical result outranks your reasoning and the finder's.
|
|
233
|
-
|
|
234
|
-
A confirmed finding ships as that failing test. Write it where its suite
|
|
235
|
-
lives, name it after the behavior it pins, and hand it over verbatim. It is
|
|
236
|
-
the brief the implementer works from and the regression test the fix keeps.
|
|
237
|
-
Run that test file alone, never its package's suite. Where a test is
|
|
238
|
-
impossible, say why in one line. "I did not write one" is not a reason.
|
|
239
|
-
|
|
240
|
-
Verdict: refuted with the disproof, or confirmed with the evidence and the
|
|
241
|
-
severity and kind you would assign. Either may differ from the finder's. Kind
|
|
242
|
-
gates what blocks the merge, so correcting a wrong kind is part of the job.
|
|
243
|
-
A class that cites a rule is checked at the citation too. Open the cited path
|
|
244
|
-
and confirm the quoted sentence is there. A quote you cannot find is dropped,
|
|
245
|
-
and your line says so. When you remain uncertain after honest effort, default
|
|
246
|
-
to refuted.
|
|
247
|
-
|
|
248
|
-
Judge each class on its own. Attempt each refutation separately and write one
|
|
249
|
-
verdict file per class. Sharing an area is a reason to read the code once,
|
|
250
|
-
never a reason to let one verdict colour another.
|
|
251
|
-
|
|
252
|
-
## Mutation brief
|
|
253
|
-
|
|
254
|
-
Run this only when the repository has a mutation testing tool for changed
|
|
255
|
-
lines. Otherwise skip it and say so in the review report.
|
|
256
|
-
|
|
257
|
-
You are the mutation checker. Your inputs are the branch diff against its
|
|
258
|
-
base, the issue, the criteria with the tests that prove them, and the instructions
|
|
259
|
-
pointer. Work only over the lines the diff adds or changes.
|
|
260
|
-
|
|
261
|
-
Work in the scratch worktree your brief names, and leave it in place. When
|
|
262
|
-
your brief names none, create one under the system temp directory, install
|
|
263
|
-
dependencies, and give its path in your report. Reviewers are reading the
|
|
264
|
-
shared tree while you run, so no mutation may touch it.
|
|
265
|
-
|
|
266
|
-
1. Run the repository's changed-lines mutation tool once over the whole diff.
|
|
267
|
-
Run it the way the repository's instructions say to run it before a push,
|
|
268
|
-
not with CI's settings. When the tool can keep its results for a later
|
|
269
|
-
run, let it, so the run after the fixes tests only what changed.
|
|
270
|
-
Every survivor it reports is a finding. Never rerun it per file, never
|
|
271
|
-
lower its budget, and never run a whole suite or a coverage run. The
|
|
272
|
-
tool can run longer than your shell's maximum timeout. Then your shell
|
|
273
|
-
moves it to the background. Wait for it to exit before you write your
|
|
274
|
-
report. Only the tool's own output can say that a run went over budget.
|
|
275
|
-
A run that ends over budget is reported as over budget with the mutants
|
|
276
|
-
it scored. One rerun is allowed, and only when the tool reports no result
|
|
277
|
-
because a workspace dependency could not be resolved: build the
|
|
278
|
-
dependencies, then run once more.
|
|
279
|
-
2. Read the tool's report. For each criterion whose production lines were
|
|
280
|
-
mutated, a test that proves it must appear among the killers of a mutant on
|
|
281
|
-
those lines. A criterion whose tests killed nothing there is a finding that
|
|
282
|
-
names the criterion as unproven. This is a read of the report, not a test
|
|
283
|
-
run.
|
|
284
|
-
3. For lines the tool never mutates, such as SQL, shell, markup or
|
|
285
|
-
configuration, break by hand each predicate, guard, cap and assignment the
|
|
286
|
-
diff introduces. Invert it, widen it, drop a conjunct, or ignore the field.
|
|
287
|
-
Judge each break by running the one test file that should catch it. If
|
|
288
|
-
nothing fails, that is a finding.
|
|
289
|
-
|
|
290
|
-
Every finding has kind proof strength and names the line, the mutation
|
|
291
|
-
applied, and the test that should have failed. Write the full report to the
|
|
292
|
-
path in your brief. Return at most ten lines: the score and whether the run
|
|
293
|
-
stayed in budget, counts by step, and the criterion ids left unproven.
|