@outerlayer/cli 0.38.0-next.20261111172503 → 0.38.0-next.20261111180956
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/build-info.json +1 -1
- package/dist/{check-HC7E2BFS.js → check-36K7RSE2.js} +5 -5
- package/dist/{chunk-DKMSDIWM.js → chunk-B5TMIK6Q.js} +1 -1
- package/dist/{chunk-YOFYW4EE.js → chunk-KIQM57WG.js} +5 -5
- package/dist/{chunk-2FJYZUZE.js → chunk-NRTT4GVU.js} +1 -1
- package/dist/{chunk-KFB7KFEK.js → chunk-OIH2WKIK.js} +1 -1
- package/dist/{chunk-KN5VDMVL.js → chunk-PGLICHNU.js} +1 -1
- package/dist/{chunk-UHOP4IMQ.js → chunk-QENFVQ7O.js} +1 -1
- package/dist/{chunk-O5ERYWXA.js → chunk-RYAEP743.js} +1 -1
- package/dist/{chunk-T6QX3S64.js → chunk-V64ULWCG.js} +3 -3
- package/dist/{chunk-LMVHXHGQ.js → chunk-VK3HNAZN.js} +2 -2
- package/dist/{cli-EOY2DOOD.js → cli-6ZLZKK2W.js} +25 -25
- package/dist/{command-OD2UYKKQ.js → command-KH62X2SL.js} +2 -2
- package/dist/{container-start-NISTFOL2.js → container-start-PELD6YZX.js} +3 -3
- package/dist/{doctor-OU54YSDJ.js → doctor-XEFEPHFI.js} +3 -3
- package/dist/{fleet-readiness-JE3MTT3R.js → fleet-readiness-BY7PY5FF.js} +1 -1
- package/dist/{host-checks-PIVAISJF.js → host-checks-67MLG6ET.js} +1 -1
- package/dist/index.js +1 -1
- package/dist/{init-CSOKCOU2.js → init-6WA65ARJ.js} +1 -1
- package/dist/{init-4WHNRYJK.js → init-RGU3LEKB.js} +1 -1
- package/dist/{init-agent-N4SCUBDT.js → init-agent-4BJN55A7.js} +1 -1
- package/dist/{init-cmd-7HDKDTUS.js → init-cmd-ZKDF2KT5.js} +2 -2
- package/dist/{init-update-UFZX37SD.js → init-update-D43GTUP3.js} +1 -1
- package/dist/{loop-DH7PXRWV.js → loop-2LVQY2G3.js} +5 -5
- package/dist/{machine-Y3VIABUB.js → machine-BW2IR5IY.js} +1 -1
- package/dist/{runner-key-DIVDANJ3.js → runner-key-6REHHUIJ.js} +1 -1
- package/dist/{status-CJSJY6XJ.js → status-TFHI7GW4.js} +1 -1
- package/dist/{update-steps-QTCJF6DY.js → update-steps-2C54XA4K.js} +2 -2
- package/dist/{work-proofs-cmd-CZFDNMXW.js → work-proofs-cmd-JVKT5OIL.js} +1 -1
- package/dist/{work-proofs-record-BVFYTMES.js → work-proofs-record-YPVLTE46.js} +1 -1
- 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
|
@@ -1,10 +1,10 @@
|
|
|
1
1
|
---
|
|
2
2
|
name: outerlayer-build
|
|
3
3
|
description: >
|
|
4
|
-
Runs the build process for one approved issue: implement it,
|
|
5
|
-
|
|
6
|
-
and
|
|
7
|
-
|
|
4
|
+
Runs the build process for one approved issue: implement it, run the
|
|
5
|
+
repository's own checks, open the pull request, and attach the criteria
|
|
6
|
+
report and the proofs as evidence. Use when the `build` skill tells you to.
|
|
7
|
+
Only the `build` skill starts it.
|
|
8
8
|
---
|
|
9
9
|
|
|
10
10
|
Run the CLI as `outerlayer`. If that command is not on PATH, run `npx -y @outerlayer/cli` with the same arguments. Every `outerlayer ...` command below follows this rule. Write the full command each time. Do not put it in a shell variable: zsh runs `$VAR args` as one command name, not as a command with arguments.
|
|
@@ -23,12 +23,11 @@ says.
|
|
|
23
23
|
This file holds the phase order and the rules that bind every phase. Each
|
|
24
24
|
phase reads its own reference on entry, never before:
|
|
25
25
|
|
|
26
|
-
- [references/
|
|
27
|
-
- [references/review-briefs.md](references/review-briefs.md): the reviewer, refuter and mutation briefs.
|
|
28
|
-
- [references/agent-briefs.md](references/agent-briefs.md): the briefs for every other agent.
|
|
26
|
+
- [references/agent-briefs.md](references/agent-briefs.md): the brief of each agent the build spawns.
|
|
29
27
|
- [references/release.md](references/release.md): sync and pull request.
|
|
30
28
|
- [references/criteria-judge.md](references/criteria-judge.md): the brief that judges each criterion.
|
|
31
29
|
- [references/code-review-judge.md](references/code-review-judge.md): the brief the runner's built-in code review follows. No phase reads it.
|
|
30
|
+
- [references/code-review-territories.md](references/code-review-territories.md): the area each reviewer of that code review searches. No phase reads it.
|
|
32
31
|
- [references/evidence.md](references/evidence.md): manual test and evidence.
|
|
33
32
|
|
|
34
33
|
Paste a brief verbatim into the agent that runs it. Add only the run's own
|
|
@@ -57,6 +56,10 @@ the step. When a team's change repeats a core step, the step runs once, and
|
|
|
57
56
|
the final report names the overlap. A change the run could not apply is
|
|
58
57
|
recorded as a finding with the `outerlayer-findings` skill.
|
|
59
58
|
|
|
59
|
+
The core has no Review phase. Do not apply a change the team lists under
|
|
60
|
+
Review. Name each one in the final report, so the team can move it to another
|
|
61
|
+
phase or remove it.
|
|
62
|
+
|
|
60
63
|
When a phase spawns agents, each agent's instructions include the team's
|
|
61
64
|
changes for that phase, pasted in. Spawned agents never read the team's
|
|
62
65
|
`build` themselves.
|
|
@@ -109,8 +112,6 @@ it does.
|
|
|
109
112
|
the model the session runs on.
|
|
110
113
|
- **Implementing model.** The model that implements. Default: the model the
|
|
111
114
|
session runs on.
|
|
112
|
-
- **Reviewing model.** The model for every review role. Default: the model the
|
|
113
|
-
session runs on.
|
|
114
115
|
- **Branch prefix.** The prefix of the branch a local build creates. Default:
|
|
115
116
|
`feat/`.
|
|
116
117
|
- **Agent budget.** The most agents one run may spawn. Default: 60.
|
|
@@ -169,9 +170,9 @@ These bind the orchestrator in every phase. Rule 5 also binds every spawned agen
|
|
|
169
170
|
tree has no changes, the gate has already passed on this code: skip the run.
|
|
170
171
|
4. **Checkpoint and re-orient.** Before the first phase, write a state file at
|
|
171
172
|
`"$(git rev-parse --absolute-git-dir)/build-state.json"`, replacing any
|
|
172
|
-
earlier one. Record the phase, the
|
|
173
|
-
agent count, the issue reference, the pull request number, the branch
|
|
174
|
-
path of the team's `build` skill
|
|
173
|
+
earlier one. Record the phase, the spawned agents, the
|
|
174
|
+
agent count, the issue reference, the pull request number, the branch and the
|
|
175
|
+
path of the team's `build` skill. Overwrite it at every phase change. On
|
|
175
176
|
entering a phase, after any wake, and after any compaction, re-orient: read
|
|
176
177
|
the state file, re-read this skill and the team's `build` skill at the path
|
|
177
178
|
it records, then read `git status`, the branch, the live agent list, and the issue or pull
|
|
@@ -182,9 +183,9 @@ These bind the orchestrator in every phase. Rule 5 also binds every spawned agen
|
|
|
182
183
|
Beside it, create `"$(git rev-parse --absolute-git-dir)/build-reports/"`,
|
|
183
184
|
emptied of any earlier run's files. Every spawned agent writes its full
|
|
184
185
|
report there and returns a summary of at most ten lines. Give each agent
|
|
185
|
-
the absolute path. Reports never land in the working tree, so
|
|
186
|
+
the absolute path. Reports never land in the working tree, so no check
|
|
186
187
|
reads them as product code. Read summaries and pass file paths, never
|
|
187
|
-
report bodies. Delete both at the end of phase
|
|
188
|
+
report bodies. Delete both at the end of phase 5.
|
|
188
189
|
Once a pull request exists, read `outerlayer work threads --item
|
|
189
190
|
"$OUTERLAYER_WORK" --json`. Each thread with `waitingOn` of `agent` holds a
|
|
190
191
|
person's comment. Answer each before the run ends, as the `outerlayer-amend` skill does.
|
|
@@ -208,7 +209,7 @@ These bind the orchestrator in every phase. Rule 5 also binds every spawned agen
|
|
|
208
209
|
green. Quoted output goes in the report to the user. Pull request bodies
|
|
209
210
|
carry curated links, never raw output.
|
|
210
211
|
7. **Files no issue.** The build files no issue anywhere, and no agent does.
|
|
211
|
-
A defect outside the change
|
|
212
|
+
A defect outside the change becomes a
|
|
212
213
|
candidate issue in a report: a title, a body and a target repository. The
|
|
213
214
|
final report lists every candidate for the user to file with `/spec`.
|
|
214
215
|
8. **Reader pass.** No text reaches the git host as the session wrote it.
|
|
@@ -220,7 +221,7 @@ These bind the orchestrator in every phase. Rule 5 also binds every spawned agen
|
|
|
220
221
|
9. **Run budget.** Spawn at most the agent budget in a run, 60 by default. Keep the count in the
|
|
221
222
|
state file. Past three quarters, say where the count stands in the status
|
|
222
223
|
line. A run that would exceed the budget stops and lists what is left
|
|
223
|
-
unverified.
|
|
224
|
+
unverified.
|
|
224
225
|
10. **Factory problems are findings.** Every agent records the problems it
|
|
225
226
|
hit outside the change with the `outerlayer-findings` skill, as its brief
|
|
226
227
|
says. The orchestrator records only the ones it hit itself. The final
|
|
@@ -358,7 +359,7 @@ Run them in order.
|
|
|
358
359
|
reading of the criteria goes under `issueReading`. The item's own list goes
|
|
359
360
|
under `recordedList`, and `differs` says whether the two lists differ.
|
|
360
361
|
|
|
361
|
-
Drafted criteria are
|
|
362
|
+
Drafted criteria are proven and judged like any others. A drafted
|
|
362
363
|
list stands like any recorded list. A later build's record
|
|
363
364
|
answers 409 until a person replaces the list with
|
|
364
365
|
`outerlayer emit criteria`, even after criteria are added to the issue. The
|
|
@@ -379,8 +380,8 @@ Run them in order.
|
|
|
379
380
|
[references/agent-briefs.md](references/agent-briefs.md), the issue, the
|
|
380
381
|
drafted criteria with their ids when the build drafted them, the branch
|
|
381
382
|
and the instructions-file pointer. The phase ends when the
|
|
382
|
-
repository's own pre-push checks and tests pass on the branch.
|
|
383
|
-
|
|
383
|
+
repository's own pre-push checks and tests pass on the branch. A branch
|
|
384
|
+
that cannot go green escalates.
|
|
384
385
|
|
|
385
386
|
The implementer runs on the Implementing model setting. Default: the model
|
|
386
387
|
the session runs on.
|
|
@@ -390,17 +391,7 @@ Run them in order.
|
|
|
390
391
|
|
|
391
392
|
The phase's gate is the Gate command setting. Default: the repository's own
|
|
392
393
|
pre-push checks and tests.
|
|
393
|
-
4. **
|
|
394
|
-
then apply the team's changes for this phase. Read [references/review-loop.md](references/review-loop.md) on
|
|
395
|
-
entry and run it. It sizes the panel to the finished diff: one generalist
|
|
396
|
-
for a small diff, the generalist and one focused reviewer for a medium
|
|
397
|
-
one, the generalist and up to three for a large or risky one. Findings are
|
|
398
|
-
merged into classes, checked by a refuter, fixed, and rechecked once. Then
|
|
399
|
-
write the review report as an HTML file and emit it:
|
|
400
|
-
|
|
401
|
-
outerlayer emit artifact <report>.html --caption "Review of the change and what it found"
|
|
402
|
-
|
|
403
|
-
5. **Pull request.** Run `outerlayer emit step "Pull request" --item "$OUTERLAYER_WORK"`,
|
|
394
|
+
4. **Pull request.** Run `outerlayer emit step "Pull request" --item "$OUTERLAYER_WORK"`,
|
|
404
395
|
then apply the team's changes for this phase. If the item fixes problems that `outerlayer findings show`
|
|
405
396
|
lists, record a result for each one before the pull request opens, with
|
|
406
397
|
`outerlayer findings result <problem id> --fixed|--not-fixed --note "<text>"`.
|
|
@@ -418,7 +409,11 @@ Run them in order.
|
|
|
418
409
|
file holds no criteria, no judge runs.
|
|
419
410
|
Do not wait for checks. Start the Evidence phase when the pull request opens,
|
|
420
411
|
as release.md says.
|
|
421
|
-
|
|
412
|
+
|
|
413
|
+
The build does not review its own change. In a runner build, the factory's
|
|
414
|
+
code review reads the pushed change with fresh agents and posts what it
|
|
415
|
+
finds on the pull request.
|
|
416
|
+
5. **Evidence.** Run `outerlayer emit step "Evidence" --item "$OUTERLAYER_WORK"`,
|
|
422
417
|
then apply the team's changes for this phase. Read [references/evidence.md](references/evidence.md)
|
|
423
418
|
when the pull request opens, and run it. Write and commit the demo test and the
|
|
424
419
|
file-proof tests. Each test uses the id of the criterion it records a proof for. When the
|
|
@@ -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.
|