@tduyng/pi-devos 0.0.0-stage → 0.14.0

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.
Files changed (109) hide show
  1. package/CHANGELOG.md +1421 -0
  2. package/LICENSE +201 -0
  3. package/README.md +206 -2
  4. package/package.json +44 -4
  5. package/plugins/devos/agents/reader.md +27 -0
  6. package/plugins/devos/agents/verifier.md +155 -0
  7. package/plugins/devos/pi/extension/devos.ts +605 -0
  8. package/plugins/devos/scripts/check-evidence-router.sh +140 -0
  9. package/plugins/devos/scripts/evidence-router.py +254 -0
  10. package/plugins/devos/scripts/gates/check-dod-core.sh +285 -0
  11. package/plugins/devos/scripts/gates/check-dod.sh +38 -0
  12. package/plugins/devos/scripts/gates/check-implementation-boundary.sh +99 -0
  13. package/plugins/devos/scripts/gates/check-port-source.sh +134 -0
  14. package/plugins/devos/scripts/gates/check-report-locations.sh +148 -0
  15. package/plugins/devos/scripts/gates/check-scenario-coverage.sh +104 -0
  16. package/plugins/devos/scripts/gates/gate-check.mjs +204 -0
  17. package/plugins/devos/scripts/gates/phase-gate.sh +208 -0
  18. package/plugins/devos/scripts/gates/plan-template-check.py +141 -0
  19. package/plugins/devos/scripts/gates/run-gate.sh +148 -0
  20. package/plugins/devos/scripts/gates/spec-contract.sh +19 -0
  21. package/plugins/devos/scripts/gates/spec-template-check.py +152 -0
  22. package/plugins/devos/scripts/gates/suite/01-dod.sh +485 -0
  23. package/plugins/devos/scripts/gates/suite/02-integrity.sh +310 -0
  24. package/plugins/devos/scripts/gates/suite/03-archetype.sh +255 -0
  25. package/plugins/devos/scripts/gates/suite/04-spec-validator.sh +183 -0
  26. package/plugins/devos/scripts/gates/suite/05-requires.sh +360 -0
  27. package/plugins/devos/scripts/gates/suite/06-clarify-adapters.sh +435 -0
  28. package/plugins/devos/scripts/gates/suite/07-lint-evals.sh +350 -0
  29. package/plugins/devos/scripts/gates/suite/08-converge.sh +591 -0
  30. package/plugins/devos/scripts/gates/suite/09-quality.sh +516 -0
  31. package/plugins/devos/scripts/gates/suite/10-eval-harness.sh +679 -0
  32. package/plugins/devos/scripts/gates/test-gates.sh +76 -0
  33. package/plugins/devos/scripts/gates/test-integrity-rubric.sh +152 -0
  34. package/plugins/devos/scripts/gates/test-phase-gate.sh +116 -0
  35. package/plugins/devos/skills/absorb/SKILL.md +75 -0
  36. package/plugins/devos/skills/absorb/references/distill.md +67 -0
  37. package/plugins/devos/skills/absorb/scripts/check-refs.sh +108 -0
  38. package/plugins/devos/skills/absorb/scripts/knowledge.py +258 -0
  39. package/plugins/devos/skills/build/SKILL.md +71 -0
  40. package/plugins/devos/skills/build/references/advanced-routing.md +19 -0
  41. package/plugins/devos/skills/build/references/defect.md +17 -0
  42. package/plugins/devos/skills/build/references/exposure.md +28 -0
  43. package/plugins/devos/skills/build/references/hard.md +39 -0
  44. package/plugins/devos/skills/build/references/refactor.md +66 -0
  45. package/plugins/devos/skills/build/references/routes.md +56 -0
  46. package/plugins/devos/skills/build/references/scale.md +84 -0
  47. package/plugins/devos/skills/build/references/spike.md +19 -0
  48. package/plugins/devos/skills/build/scripts/construct-census.py +204 -0
  49. package/plugins/devos/skills/build/scripts/dep-order.py +297 -0
  50. package/plugins/devos/skills/build/scripts/ledger.sh +228 -0
  51. package/plugins/devos/skills/build/scripts/route.sh +5 -0
  52. package/plugins/devos/skills/build/scripts/secret-scan.sh +121 -0
  53. package/plugins/devos/skills/build/scripts/toolchain-preflight.py +302 -0
  54. package/plugins/devos/skills/frame/SKILL.md +54 -0
  55. package/plugins/devos/skills/frame/references/template.md +115 -0
  56. package/plugins/devos/skills/frame/scripts/survey.sh +44 -0
  57. package/plugins/devos/skills/implement/SKILL.md +36 -0
  58. package/plugins/devos/skills/implement/references/conditional-plan.md +17 -0
  59. package/plugins/devos/skills/implement/references/debug-loop.md +7 -0
  60. package/plugins/devos/skills/implement/references/dod.md +37 -0
  61. package/plugins/devos/skills/implement/references/implement.md +54 -0
  62. package/plugins/devos/skills/implement/references/manual-gates.md +16 -0
  63. package/plugins/devos/skills/implement/references/plan.md +142 -0
  64. package/plugins/devos/skills/implement/references/resume.md +9 -0
  65. package/plugins/devos/skills/implement/references/scale.md +86 -0
  66. package/plugins/devos/skills/implement/references/verify-close.md +62 -0
  67. package/plugins/devos/skills/implement/references/visual-verification.md +7 -0
  68. package/plugins/devos/skills/implement/scripts/check-converge.sh +147 -0
  69. package/plugins/devos/skills/implement/scripts/equivalence-check.sh +140 -0
  70. package/plugins/devos/skills/implement/scripts/manifest-queue.sh +185 -0
  71. package/plugins/devos/skills/implement/scripts/ops-log.sh +58 -0
  72. package/plugins/devos/skills/implement/scripts/pilot-consistency.py +173 -0
  73. package/plugins/devos/skills/implement/scripts/validate-spec.sh +220 -0
  74. package/plugins/devos/skills/init/SKILL.md +100 -0
  75. package/plugins/devos/skills/init/references/archetype-sections.md +15 -0
  76. package/plugins/devos/skills/init/references/context.md +42 -0
  77. package/plugins/devos/skills/init/references/conventions-template.md +153 -0
  78. package/plugins/devos/skills/init/references/reverse-engineer.md +3 -0
  79. package/plugins/devos/skills/init/references/rule-format.md +58 -0
  80. package/plugins/devos/skills/init/references/shape-model.md +29 -0
  81. package/plugins/devos/skills/init/scripts/detect-archetype.py +461 -0
  82. package/plugins/devos/skills/init/scripts/render-rule.py +114 -0
  83. package/plugins/devos/skills/init/scripts/reverse-engineer.sh +162 -0
  84. package/plugins/devos/skills/learn/SKILL.md +29 -0
  85. package/plugins/devos/skills/learn/references/destinations.md +43 -0
  86. package/plugins/devos/skills/learn/references/scoring.md +26 -0
  87. package/plugins/devos/skills/review/SKILL.md +54 -0
  88. package/plugins/devos/skills/review/references/comments.md +13 -0
  89. package/plugins/devos/skills/review/references/gaps.md +26 -0
  90. package/plugins/devos/skills/review/references/reviewer.md +124 -0
  91. package/plugins/devos/skills/review/references/routing.md +35 -0
  92. package/plugins/devos/skills/review/scripts/__pycache__/prepare.cpython-313.pyc +0 -0
  93. package/plugins/devos/skills/review/scripts/prepare.py +763 -0
  94. package/plugins/devos/skills/spec/SKILL.md +57 -0
  95. package/plugins/devos/skills/spec/references/prototype.md +21 -0
  96. package/plugins/devos/skills/spec/references/question-craft.md +47 -0
  97. package/plugins/devos/skills/spec/references/roadmap-link.md +5 -0
  98. package/plugins/devos/skills/spec/references/write-spec.md +128 -0
  99. package/plugins/devos/skills/spec/scripts/clarify.sh +173 -0
  100. package/plugins/devos/skills/status/SKILL.md +67 -0
  101. package/plugins/devos/skills/status/references/cost.md +38 -0
  102. package/plugins/devos/skills/status/scripts/artifact-budgets.sh +88 -0
  103. package/plugins/devos/skills/status/scripts/check-requires.py +556 -0
  104. package/plugins/devos/skills/status/scripts/context-cost.sh +89 -0
  105. package/plugins/devos/skills/status/scripts/convention-usage.sh +184 -0
  106. package/plugins/devos/skills/status/scripts/findings-summary.sh +112 -0
  107. package/plugins/devos/skills/status/scripts/quality-correlation.sh +259 -0
  108. package/plugins/devos/skills/status/scripts/sizing-calibration.sh +71 -0
  109. package/plugins/devos/skills/status/scripts/spec-diff.sh +137 -0
package/CHANGELOG.md ADDED
@@ -0,0 +1,1421 @@
1
+ # Changelog
2
+
3
+ ## 0.14.0: 2026-11-16
4
+
5
+ ### Added
6
+ - Report the settled findings the code no longer shows
7
+ - --comments drafts answers to a teammate's review comments
8
+
9
+ ### Changed
10
+ - Keep the frame skill smaller, generate only one markdown file
11
+ - Build the packet in a script, hand the reviewer its rubric
12
+ - Gates read the contract from spec.md, legacy spec-functional.md still accepted
13
+ - One spec.md carries contract and plan, docs and skills drop spec-functional.md
14
+ - Plan appends to spec.md below the contract, one Technical Requirements heading
15
+ - Drop the spec-functional.md fallback, fixtures and evals use spec.md only
16
+ - Out of scope lives in the contract only, Boundaries drops its Never tier
17
+ - Drop the team handoff gate, devos is solo
18
+ - Notes hold Q&A and assumptions only, anchors are written once in spec.md
19
+ - Split test-gates.sh into suite/ parts
20
+ - Plan.md drops the restated gate list and anchor bullet
21
+ - Drop the docs and product skills and their workflows
22
+ - Drop the grep cookbook and rationalization table
23
+ - Run the test-gates suite parts in parallel, each with its own workspace
24
+ - Fold the discipline skills into implement, drop design-taste and explain, write pr.md at close
25
+
26
+ ## 0.13.6: 2026-11-11
27
+
28
+ ### Added
29
+ - Add the frame stage and absorbed product knowledge
30
+ - Audit the ask and widen the impact dimensions
31
+ - Name the shape you rejected and why
32
+ - Read the product knowledge before planning
33
+ - Check the knowledge base before quoting it
34
+ - Keep drift signal from drowning in out-of-scope sources
35
+ - Report a spent spike as finished
36
+ - Ship the page-to-code reference check
37
+
38
+ ### Changed
39
+ - Exercise the frame survey script
40
+
41
+ ### Documentation
42
+ - Document absorb and the knowledge files, point init at the manifest script
43
+ - Audit the request for the four things it skipped
44
+ - A frame ends the run that wrote it
45
+ - Say how a project wires itself to a clone
46
+ - A collection page is not the whole collection
47
+ - Say how often the frame may ask
48
+ - Put absorb and frame on the front door
49
+
50
+ ### Fixed
51
+ - Resolve the skill directory outside Claude
52
+ - Count a reimport as nothing when nothing changed
53
+ - A spike folder is finished, not a spec missing its plan
54
+
55
+ ## 0.13.5: 2026-11-10
56
+
57
+ ### Added
58
+ - Hard route and plan gates so smaller models cannot skip phases
59
+ - Route and plan gates on Claude Code, Cursor, Codex, Copilot and opencode
60
+
61
+ ## 0.13.4: 2026-11-08
62
+
63
+ ### Added
64
+ - Restore cursor support as a cursor plugin
65
+ - Block DONE on a scenario the report never records
66
+
67
+ ### Fixed
68
+ - Pass verify on BSD grep, BSD sed, and pi 1.5.0
69
+
70
+ ## 0.13.3: 2026-11-02
71
+
72
+ ### Added
73
+ - Bind the verification report to the commit it judged
74
+ - Name the blind spot and the question a human must answer
75
+ - Name the shared premise, test the returns-nothing case
76
+
77
+ ### Documentation
78
+ - Verify the Pi adapter against pi 1.4.0
79
+ - Fix the Pi skill command typo
80
+ - Name the end-of-turn gate for each harness
81
+ - Refresh the codebase map and conventions to 1744ba4
82
+ - The route entity lists all seven routes
83
+
84
+ ## 0.13.2: 2026-10-26
85
+
86
+ ### Added
87
+ - The verifier runs its real rubric and every path field expands
88
+ - Read the plan and the diff in a fresh context
89
+
90
+ ### Changed
91
+ - Record the port scenarios' honest scores
92
+ - Record the port scenarios' honest scores
93
+ - DevOS ships no agent dispatch and no model choice
94
+ - Drop the System One pass from the evidence router
95
+
96
+ ### Fixed
97
+ - A skill's script paths and adapter tool names resolve on an installed project
98
+
99
+ ## 0.13.1: 2026-10-21
100
+
101
+ ### Added
102
+ - S route offers learn before it stops
103
+ - A scale path for behavior-preserving changes
104
+ - Measure the C route at scale against a real model
105
+ - The port unit order comes from the tree, not from a guess
106
+ - The port's decisions come from a census of the source
107
+ - Measure the port order and the cycle against a real model
108
+ - The port asks what the tree still carries, and the judge has to see a break
109
+ - The port pilots three units and prices the judgment from a measurement
110
+ - A unit ports its test first, and an upgrade reads the release notes it ships with
111
+ - The exposure answer is a scan and a reading, in one reference
112
+ - Measure the port's judgment discipline, and name the port's close
113
+ - A port whose rules do not fit a capsule is a port in phases
114
+ - A phase of a port is a slice of the manifest and a census of that slice
115
+
116
+ ### Changed
117
+ - The amendment's cost is the census count, and the judge rule says it once
118
+ - Record the port scenarios' honest scores
119
+
120
+ ### Documentation
121
+ - The repo map names the artifacts a wide change leaves behind
122
+ - The census has a spec, and two task groups run in order
123
+
124
+ ### Fixed
125
+ - Use a portable in-place sed in two evaluation scripts
126
+ - Ship the Codex marketplace manifest
127
+ - The on-invoke cost ratchet never ran for DevOS's own skills
128
+ - Report the session cost the harness actually loads
129
+ - The template cache must hold every scenario
130
+ - None with a reason is still None, and a port's capsule names a directory
131
+ - The rules reach a worker two ways, and the rulebook stays short
132
+ - The rulebook has to fit the capsule, and the capsule is how a worker gets it
133
+ - A long import chain does not walk off the interpreter's stack
134
+ - The scan prunes vendored trees, and a manifest row with no target is a finding
135
+ - A port's rule is scoped to the file it governs, and it restates at the close
136
+ - A criterion that names a script must see the command, not the reading
137
+ - A criterion sees the whole command, and the ported tree's tests run as they are written
138
+ - A declared guard has to hold before any agent runs
139
+ - A port is one behavior over many files, and C is its route
140
+ - The scale threshold is a file count, because a count can be checked
141
+ - The user asking for the work to be tracked selects the scale path too
142
+
143
+ ## 0.13.0: 2026-10-02
144
+
145
+ ### Added
146
+
147
+ - Ground implementation in local craft
148
+ - Check the plan.md template against the gates, group Phase A into 4 steps
149
+ - Gate unpushed commits on the conventional-commit grammar
150
+ - Check the M spec shape against validate-spec, put it before the L template
151
+ - Add a script census that separates logic from data
152
+ - Teach the doc-drift gate about the check-requires migration
153
+ - Integrate with laya
154
+ - Run laya by default and log what it disagreed with
155
+
156
+ ### Changed
157
+
158
+ - Cover minimal algorithm choice
159
+ - Reduce always-loaded descriptions
160
+ - Remove redundant DevOS prefixes
161
+ - Remove repeated trigger boilerplate
162
+ - Reduce always-loaded descriptions
163
+ - Compress the three phase files to rules without essays
164
+ - Write the functional spec in the interview turn instead of a subagent
165
+ - One question round and one approval gate on the spec pipeline
166
+ - Trim the workspace and quality-row prose
167
+ - Fold the content-gap audit into a synthesis checklist
168
+ - Name a one-page M spec shape and treat 300 lines as a ceiling
169
+ - Move the parallel-marker essay out of the plan template
170
+ - Fold the advanced-routing triggers into the router
171
+ - Put the strange-request rule on the S path
172
+ - Cut the quality-row instruction to the column list
173
+ - List the conventions rules instead of one long paragraph
174
+ - Trim the review brief and the deletion-test restatement
175
+ - Guard that a one-page M spec folder closes the pipeline
176
+ - Group Phase C into 5 steps and scale the close to the change
177
+ - Group Phase B into 4 steps and scale it to the change
178
+ - Put the S test and the S route ahead of the routing table
179
+ - Make planning/notes.md the interview record, not a second spec
180
+ - Fold user stories into each functionality subsection, gate both shapes
181
+ - Drop the duplicated validate-spec step, trigger decisions-input explicitly
182
+ - Write a six-column quality row, gate that the parser counts it
183
+ - Cut the self-trigger argument and two design essays in learn
184
+ - Fold two thin references into write-spec, make the prototype skip decidable
185
+ - State the gate rejections so the planner need not open the scripts
186
+ - Port check-requires to python, 587x faster on --concurrent
187
+ - Add just probe and just census for the unwired audit tools
188
+ - Port slop-scan to python, 14x faster on one file
189
+ - Extract the score heredocs into an importable module
190
+ - Port render-rule to python, 12 processes per rule file
191
+ - Port detect-archetype to python, 14x faster
192
+ - Port check-doc-drift to python, 43x faster
193
+ - Split harness-depth checks out of the per-commit gate
194
+ - Rename the laya pass to System One and make the provider a setting
195
+ - Simplify pi extension
196
+
197
+ ### Documentation
198
+
199
+ - Match the simplified skills, and gate against drift
200
+ - Update docs
201
+
202
+ ### Fixed
203
+
204
+ - Keep process metadata out of implementation output
205
+ - Serialize template cache rebuilds
206
+ - Parallelize independent template builds
207
+ - Bound stubborn tool timeouts
208
+ - Reject blank quality audit fields
209
+ - Reject malformed quality audit components
210
+ - Take findings trend month from the record field
211
+ - Require a first-pass claim for zero-findings credit
212
+ - Report task groups converge cannot identify
213
+ - Keep converge findings on one line
214
+ - Report parallel batches naming an undefined group
215
+ - Compare plan overview count to task groups
216
+ - Accept both markdown bullet forms in convention usage
217
+ - Read task groups with the plan-wide header shape in spec diff
218
+ - Read a size call only from a letter that stands alone in sizing calibration
219
+ - Read the package manager from bun text lockfile in stack facts
220
+ - Say when a capped ledger write dropped content
221
+ - Count an empty frontend file in slop-scan instead of calling it binary
222
+ - Tell a non-repo apart from an empty history in reverse-engineer
223
+ - Read a colon-form task group as a group in the plan gates
224
+ - Compare a colon-form task group in the spec structural diff
225
+ - Match a directory-carrying archetype glob with find -path
226
+ - Search a convention capsule whose spec directory name contains a space
227
+ - Print the quality-correlation route tables in declared S/M/L/C order
228
+ - End a Requires section on any heading, not only a lettered one
229
+ - Reject a rule whose frontmatter never closes
230
+ - Strip the Requires label the same way the reader matches it
231
+ - Gate unresolvable and orphaned references, drop three dead files
232
+ - Move form guidance out of the spec blocks and gate against it leaking back
233
+ - Restore two scenario docs whose edits desynchronised them from the machine criteria
234
+ - Repoint clarify's acceptance-criterion check at the folded spec shape
235
+ - Extend doc drift to the skills, and resolve bare script names
236
+ - Bound the template build lock wait and name the holder
237
+
238
+ ## 0.12.0: 2026-09-15
239
+
240
+ ### Added
241
+
242
+ - Surface human callouts, calibrate findings, drop stale ones
243
+ - Simplify verify step, rewrite plugins installation and remove cursor
244
+ - Restructure plugin tools name
245
+ - Support only opencodeV2, not v1
246
+ - Remove generated tools and scripts
247
+
248
+ ### Changed
249
+
250
+ - Consolidate the three plugins into one devos plugin
251
+ - Reduce to one subagent profile (verifier)
252
+ - Merge spec-write into spec
253
+ - Fold claude-context and domain-model into onboard
254
+ - Merge specs, stats, and diff into status
255
+ - Fold verify-gap into review as --gaps
256
+ - Merge the doc commands into docs and auto-onboard in build
257
+ - Update setup evals
258
+ - Setup devos
259
+ - Rename onboard to init
260
+ - Rename product-plan to product
261
+ - Drop the authority flag, use disable-model-invocation
262
+ - Default to a [spec-name] branch and .worktrees/
263
+ - Rename pi-extension to pi-devos
264
+ - Remove roadmap folder
265
+ - Publish only necessary files
266
+ - Update setup
267
+ - Remove the orphaned post-merge hook and stale symlink comments
268
+ - Wire adopt-check into a recipe and fix a stale AGENTS reference
269
+ - Update pi docs
270
+ - Simplify all skill descriptions
271
+ - Update evals score
272
+
273
+ ### Documentation
274
+
275
+ - Add the DevOS command surface redesign
276
+ - Rewrite the front door with a lifecycle diagram and worked example
277
+ - Drop stale plugin and multi-agent references
278
+ - Record the implemented redesign and its deviations
279
+ - Restore the full route schema and rename stale eval scenarios
280
+ - Remove internal pages and fix support-matrix counts
281
+ - Remove troubleshooting page and its links
282
+ - Shorten configuration and team workflow, drop glossary and situations
283
+ - Remove security, support-matrix, and examples consumers
284
+ - Trim configuration, concepts, and command pages to essentials
285
+ - State why product and docs are optional
286
+ - Remove headless doc
287
+ - Drop per-tool agent references after the generator removal
288
+ - Copilot invokes skills by name, no command wrappers or agent
289
+ - Drop references to removed docs and stale installer comment
290
+ - Refresh the codebase map and the Codex README after the cleanup
291
+ - Pi and Codex no longer ship a verifier agent
292
+ - Drop the Roadmap/Tier/Rule entities and stale conventions
293
+ - Align command syntax, flags, and verifier note with the skills
294
+ - Use GitLab URLs for the Copilot and OpenCode remote installs
295
+ - Drop the last references to the removed roadmaps folder
296
+ - Clear about role member
297
+
298
+ ### Fixed
299
+
300
+ - Check dod score
301
+ - Adapt opencode installer and adapter docs to one plugin
302
+ - Quote the skill description so YAML frontmatter parses
303
+ - Reconcile adapter symlinks so reinstall prunes stale entries
304
+ - Explicite string for skill description
305
+ - Correct plugin paths
306
+ - Close four fail-open gates and correct the install docs
307
+
308
+ ## 0.11.9: 2026-09-12
309
+
310
+ ### Added
311
+
312
+ - Add the H route for fresh-context decomposition
313
+
314
+ ## 0.11.8: 2026-09-10
315
+
316
+ ### Added
317
+
318
+ - Update worktrees rules
319
+
320
+ ### Fixed
321
+
322
+ - Ship the delegate tool by default, discover symlinked agents, map /devos: follow-ups
323
+ - Stop a watchdog subshell from firing the caller's EXIT cleanup
324
+
325
+ ## 0.11.7: 2026-09-09
326
+
327
+ ### Added
328
+
329
+ - Fix missing suggest next step
330
+
331
+ ### Documentation
332
+
333
+ - Add install by project guide
334
+ - Document the per-machine agent override
335
+
336
+ ### Fixed
337
+
338
+ - Wrap /skill: invocations so onboard runs instead of asking
339
+
340
+ ## 0.11.6: 2026-09-05
341
+
342
+ ### Fixed
343
+
344
+ - Probe the gate that exists, make a bare onboard scan instead of asking
345
+ - Catch a stale plugin snapshot at an unchanged version
346
+ - Backfill mandatory conventions sections into an existing file
347
+
348
+ ## 0.11.5: 2026-09-05
349
+
350
+ ### Added
351
+
352
+ - Recommend the smallest option, read the nearest exemplar, detect CI topology
353
+ - Gate the capsule before approval, carry overrides into the spec
354
+ - Fail a plan whose additions answer no failed item
355
+ - Read every runner's checks, detect containers, run checks where the project runs them
356
+ - Poll async work to terminal state, never infer pass from kickoff
357
+
358
+ ### Changed
359
+
360
+ - Improve install command
361
+
362
+ ## 0.11.4: 2026-09-03
363
+
364
+ ### Added
365
+
366
+ - Probe real need and lifecycle states before freezing the spec
367
+ - Falsifiable hypothesis in diagnosis, deletion test for new modules
368
+
369
+ ## 0.11.3: 2026-09-01
370
+
371
+ ### Added
372
+
373
+ - Prewalk-informed cost and closure changes
374
+
375
+ ### Fixed
376
+
377
+ - Rearm pre-commit as the adapter-drift check
378
+
379
+ ## 0.11.2: 2026-09-01
380
+
381
+ ### Changed
382
+
383
+ - Fix skill syntax for pi
384
+ - Regenerate adapters for quoted yaml descriptions
385
+
386
+ ## 0.11.1: 2026-08-31
387
+
388
+ ### Added
389
+
390
+ - Set pi as default, rerun and fix some criteria
391
+ - Baseline cheater probe, 16/32 scenarios pass
392
+ - Ground transcript criteria to filesystem
393
+ - Add word_soup_leak metric, close the last soup leaks
394
+ - Delegate to a chosen model and optional lean pass
395
+
396
+ ### Changed
397
+
398
+ - Golden transcripts for the 8 scenarios without oracles
399
+ - Raise oracle coverage from 7 to 15 scenarios
400
+ - Raise oracle coverage from 7 to 15 scenarios
401
+ - Add oracles for 05b/05c/05d/12/19, fix 19 fixture missing require
402
+ - Add oracles for 03/04/07, fix test-location mismatch
403
+ - Add oracles for 06c/06d, align 06d criterion to scenario
404
+ - Add oracles for 02/06b/20, fix parallel-marker and test-location bugs
405
+ - Complete oracle coverage, all 32 scenarios gated
406
+
407
+ ### Fixed
408
+
409
+ - Criterion and fixture bugs behind 15/17/06g floors
410
+ - Single DoD prefix, name spec error for bare archive path
411
+ - Ellipsis and pasted report path are missing-spec, not missing report
412
+ - Ground 06i, 08, 10, 21 to filesystem artifacts
413
+ - Ground 16 and 08, word_soup_leak stays 0
414
+ - Defeat natural-language pairing cheater
415
+ - Block destructive-cheater hole
416
+ - Close artifact-fabrication cheat
417
+ - 16 phases-reported rejects all wrong phase pairings, not just gamma-complete
418
+ - Reject wrong-direction and negated claims in 08/16
419
+ - Serialize template-cache rebuild with a lock
420
+ - Close toucher gain on 16 via ordered blocked-by chain
421
+ - Enforce must_hold guardrails with a hard exit in devos mode
422
+ - Enforce must_flip discrimination guard with a hard exit in baseline mode
423
+ - Lock in must_flip/must_hold exit-code enforcement with selftests
424
+ - Empty transcript is UNMEASURED, not a precondition score
425
+ - Cover transcript.py's pi schema in the oracle gate
426
+ - Lock multi-turn pi transcript block isolation with selftests
427
+ - Named-evidence resolves ambiguous bare names to any matching file
428
+ - Named-evidence bare-name resolution is case-insensitive
429
+ - Said-near gap is inclusive at the boundary
430
+ - 08 fixture lacked the user stories its prompt demands
431
+ - 03 scores spec-not-overwritten, rebalanced so no-op stays below 50%
432
+ - 18's parallel dispatch was unscored; baseline scored 100%
433
+ - 13-c rename-complete had a corrupted regex (vacuous check)
434
+ - Bare .eval-base in git diff made 10 negative checks vacuous
435
+ - Lock run_with_timeout behavior with selftests
436
+ - 15's done-requires-evidence accepted the stale fixture report
437
+ - 20's plan-recreated was satisfiable by a fabricated empty spec.md
438
+
439
+ ## 0.11.0: 2026-08-26
440
+
441
+ ### Added
442
+
443
+ - Greenfield support - build routes normally on empty repos, onboard skips mapping
444
+ - Preset packs - render-rule accepts directories, presets doc
445
+ - Remove cross-review skill, code-reviewer agent, and all references
446
+ - One-command installer for all 6 tools with dry-run detection
447
+ - Scenario traceability gate - validate-spec cross-checks SC-n coverage between functional spec and plan capsule
448
+ - Implementation notes propagation in capsule; SC-n ids stay out of source comments; refresh measured token rows
449
+ - Rule supersession via frontmatter, render-rule skips dead rules
450
+ - Evolve pass over eval results history
451
+ - Clean-context debug escalation at second non-converging verifier round
452
+ - Must_flip criteria flag with baseline discrimination guard
453
+ - Make pi first principle
454
+ - Self tune
455
+ - Install.sh reports per-tool failures, wire into justfile and README
456
+ - Just update wraps git pull + Claude Code snapshot refresh
457
+
458
+ ### Changed
459
+
460
+ - Improve evals performance
461
+ - First pi eval runs, devos vs baseline on 01-small-fix (75% both)
462
+ - Add check-converge
463
+ - Hold criteria flag
464
+ - Run verification clean pass
465
+ - Prove pi install git:/local-path route against the real pi binary
466
+ - Rename-complete to scenario 13
467
+ - Adapters/pi -> pi-extension, drop the generic adapter wrapper for Pi
468
+ - Diagnostic scenario 19
469
+ - Drop adapters/ wrapper, each tool lives in its native top-level folder
470
+ - Unify symlink installers into data-driven script
471
+ - Add hook-command-path resolution guard
472
+ - Collapse gate shims into data-driven run-gate.sh
473
+ - Add block-probe to check-support-matrix.sh
474
+ - Simplify docs
475
+ - Fix lint scripts
476
+
477
+ ### Documentation
478
+
479
+ - Support-matrix updated
480
+ - Three induced failure modes from pi eval sessions
481
+ - Redgreen authoring protocol, errored-eval-score failure mode
482
+
483
+ ### Fixed
484
+
485
+ - Drop pinned external model IDs from code-reviewer, CLIs track their current default
486
+ - A cwd bug
487
+ - Build router - multi-behavior requests route at least M (pi free-model trial evidence)
488
+ - Errored eval runs score UNMEASURED, never from partial transcripts
489
+ - Scope skill-cost baseline lint to core plugins, not third-party extensions
490
+ - Repair multi-agent install layer
491
+ - Warn gate payload and update docs
492
+
493
+ ## 0.10.16: 2026-08-24
494
+
495
+ ### Added
496
+
497
+ - Optimise token efficient with autoreseearch
498
+
499
+ ## 0.10.15: 2026-08-23
500
+
501
+ ### Added
502
+
503
+ - Ops trail, one JSONL line per phase transition
504
+ - Promote dead ends and behavior conflicts at close, not just findings
505
+
506
+ ## 0.10.14: 2026-08-22
507
+
508
+ ### Added
509
+
510
+ - Register ask_user tool for structured spec-interview questions
511
+ - Acceptance gates, progress-aware pi correction budget
512
+
513
+ ### Changed
514
+
515
+ - Plan approval in conversation, drop harness plan-mode coupling
516
+ - Drop provenance notes from gates work
517
+ - Runner only, drop DoD re-parse of gates.md until a real spec uses one
518
+
519
+ ### Documentation
520
+
521
+ - Full 31-scenario pi run on x-preview-f-free, all at or above floor
522
+
523
+ ### Fixed
524
+
525
+ - Read pi transcripts so transcript criteria score instead of falsely failing
526
+
527
+ ## 0.10.13: 2026-08-21
528
+
529
+ ### Added
530
+
531
+ - Support devos as pi entension
532
+
533
+ ### Documentation
534
+
535
+ - Add Project Context section to AGENTS.md so plain sessions see the artifacts
536
+ - Refresh codebase map to c3f918df
537
+ - Document pi's two-layer CLAUDE_SKILL_DIR resolution
538
+
539
+ ### Fixed
540
+
541
+ - Persist gate correction budget in session entries so reload cannot re-spend it
542
+
543
+ ## 0.10.12: 2026-08-19
544
+
545
+ ### Added
546
+
547
+ - Skill learn can invoke by agent
548
+
549
+ ### Fixed
550
+
551
+ - Recognize it.each/test.each/describe.each as consolidation
552
+ - Reword learn description and update stale token-count docs
553
+
554
+ ## 0.10.11: 2026-08-17
555
+
556
+ ### Added
557
+
558
+ - Rewrite run.sh script, make it argnostic run, no depend to claude
559
+
560
+ ### Changed
561
+
562
+ - Establish per-scenario scoring floors from 2026-08-17 run
563
+ - Run 9 cap-limited scenarios via pi
564
+ - Correct floor values in pi results
565
+ - Update cross-skill lint
566
+
567
+ ## 0.10.10: 2026-08-15
568
+
569
+ ### Added
570
+
571
+ - Enhance guardrails
572
+
573
+ ## 0.10.9: 2026-08-14
574
+
575
+ ### Added
576
+
577
+ - Provenance gates a lesson before durability does
578
+ - Every acceptance criterion carries its own evidence row
579
+ - Findings.md carries what it could not ground
580
+ - Walk back the work the pipeline kept out of your context
581
+
582
+ ## 0.10.8: 2026-08-13
583
+
584
+ ### Added
585
+
586
+ - Simplify opencode adapter
587
+ - Rewrite adapter match opencode2 rules
588
+ - Artifact budget bands enforced in just verify
589
+ - Convention usage report gives conventions.md a prune signal
590
+ - Budget band decides what a write to a capped artifact may do
591
+ - Abandoned approaches get a rejected: decisions entry
592
+
593
+ ### Changed
594
+
595
+ - Domain-model back under its 60-line page budget
596
+ - Drop mine-conventions.sh, unwired and unable to cluster prose
597
+
598
+ ### Documentation
599
+
600
+ - Architecture.md states the map's real 120-line cap
601
+
602
+ ### Fixed
603
+
604
+ - Drop non-devos command wrappers from generated adapter dir
605
+
606
+ ## 0.10.7: 2026-08-11
607
+
608
+ ### Added
609
+
610
+ - Make compatible with claude env
611
+ - Make harness agnostic and fix pi subagent behavior
612
+
613
+ ### Fixed
614
+
615
+ - Install pi better on existing settings
616
+
617
+ ## 0.10.6: 2026-08-11
618
+
619
+ ### Added
620
+
621
+ - Improve evals gates
622
+ - Support piagent and improve stats skill
623
+ - Improve devos for pi agent
624
+ - Fix pi agent scripts
625
+
626
+ ### Changed
627
+
628
+ - Switch to Apache-2.0
629
+ - Fix skills syntax for pi agent
630
+
631
+ ### Documentation
632
+
633
+ - Simplify some page docs
634
+ - Simplify docs, keep shorter
635
+ - Update docs for pi agent
636
+
637
+ ### Fixed
638
+
639
+ - Evals issues
640
+
641
+ ## 0.10.5: 2026-08-10
642
+
643
+ ### Added
644
+
645
+ - Design-taste discipline, anti-slop frontend with a scanner
646
+
647
+ ### Changed
648
+
649
+ - Raise the negative-fixture floor to 10
650
+
651
+ ### Documentation
652
+
653
+ - Route design-taste through the same doc surfaces every skill uses
654
+
655
+ ## 0.10.4: 2026-08-09
656
+
657
+ ### Added
658
+
659
+ - Attach specialist capability instead of absorbing it
660
+
661
+ ### Fixed
662
+
663
+ - Fix promotion script
664
+
665
+ ## 0.10.3: 2026-08-08
666
+
667
+ ### Added
668
+
669
+ - Rerun evals for all skills
670
+
671
+ ### Changed
672
+
673
+ - Close all checkboxes
674
+
675
+ ## 0.10.2: 2026-08-08
676
+
677
+ ### Added
678
+
679
+ - Improve review skills
680
+ - Contradiction is a category, not something the reader notices
681
+ - The C route, plus the fence and grouping it exposed
682
+
683
+ ### Documentation
684
+
685
+ - ROADMAP_023, the framework only knew how to add behavior
686
+ - The correlation report grew two sections, say so
687
+ - The 06-series sweep is deferred, not open work with no owner
688
+ - First real C-route run, cheap and honest about what it found
689
+ - The ∥-dispatch question is answered, weak prescription confirmed on Sonnet too
690
+
691
+ ### Fixed
692
+
693
+ - The 01-05 series ran clean, the 06 series ran into a quota wall
694
+ - C route now hands off to /devos:implement, proven on a real run
695
+
696
+ ## 0.10.1: 2026-08-06
697
+
698
+ ### Added
699
+
700
+ - Hedge words and code blocks are checked, not requested
701
+
702
+ ### Changed
703
+
704
+ - Shape-model.md drops the puppetry, 183 lines to 29
705
+
706
+ ### Documentation
707
+
708
+ - Why spec-functional.md and spec.md stay two files
709
+ - The embed rule and notes.md are load-bearing, not legacy
710
+
711
+ ## 0.10.0: 2026-08-06
712
+
713
+ ### Added
714
+
715
+ - Improve evals
716
+ - Converge on the smallest design before writing the plan
717
+ - A test run that executed nothing is UNVERIFIED, not PASS
718
+ - Gate per-moment load claims (check-claims check 4)
719
+ - The run transcript becomes scoreable evidence
720
+ - The pre-change result is captured, so "pre-existing" is provable
721
+ - Command output is data too, and 06h tests it
722
+
723
+ ### Documentation
724
+
725
+ - Update docs
726
+ - Re-measure the implement phase-load figures (rule 6)
727
+ - One canonical index, duplication cut from every roadmap
728
+ - Compress every roadmap to a changelog, drop unspent-eval tracking
729
+ - 022 closes, with the finding it was not looking for
730
+
731
+ ### Fixed
732
+
733
+ - Improve score calculation
734
+ - .quality.md recorded task-group count as the route
735
+ - The routing analyzer lives where the user's data is
736
+ - The scoring floor moves on the trial median, not the luckiest run
737
+ - The harness could never complete a run, so every floor read 0%
738
+ - Drop --add-dir for the plugin, and the injection report now fires
739
+
740
+ ## 0.9.8: 2026-08-03
741
+
742
+ ### Changed
743
+
744
+ - Move the B and R routes to references/, gate on-invoke cost per skill
745
+
746
+ ### Documentation
747
+
748
+ - Roadmap 021 status, tiers 1-5 are in 0.9.7 and tier 6 is not
749
+
750
+ ## 0.9.7: 2026-08-03
751
+
752
+ ### Added
753
+
754
+ - A defect needs a red-capable command before a hypothesis
755
+ - Give the injection-defense claim a runner, and reach the main agent with it
756
+
757
+ ### Changed
758
+
759
+ - Clean docs
760
+
761
+ ### Documentation
762
+
763
+ - Make docs more clear
764
+ - Add roadmap 21
765
+ - Simplify roadmap 21
766
+ - Roadmap 021, the ops layer rejected and what replaced it
767
+
768
+ ## 0.9.6: 2026-08-01
769
+
770
+ ### Added
771
+
772
+ - Enhance token efficiency
773
+
774
+ ### Documentation
775
+
776
+ - Add docs for devos:learn command
777
+
778
+ ## 0.9.5: 2026-08-01
779
+
780
+ ### Added
781
+
782
+ - A checkable lesson becomes a check, not a bullet
783
+
784
+ ### Documentation
785
+
786
+ - Roadmap 020, the lesson that becomes a check
787
+
788
+ ## 0.9.4: 2026-07-31
789
+
790
+ ### Fixed
791
+
792
+ - Inject the skill into the command, don't ask for it
793
+
794
+ ## 0.9.3: 2026-07-30
795
+
796
+ ### Added
797
+
798
+ - Name the design axes where unspecified has a default
799
+
800
+ ### Documentation
801
+
802
+ - Fix all docs for tiers 40-44 (B route, rule 25, Preview command)
803
+ - Tiers 45 and 46
804
+
805
+ ## 0.9.2: 2026-07-30
806
+
807
+ ### Added
808
+
809
+ - Design direction is decided once, not per feature
810
+
811
+ ## 0.9.1: 2026-07-30
812
+
813
+ ### Added
814
+
815
+ - The app is started and looked at
816
+
817
+ ## 0.9.0: 2026-07-30
818
+
819
+ ### Added
820
+
821
+ - Resolved stack versions reach the capsule
822
+ - Every check the project declares is run and reported by name
823
+ - A defect has a route
824
+
825
+ ### Documentation
826
+
827
+ - Add roadmap 019
828
+ - Tick roadmap 019 tiers 40, 41, 44 and re-measure invoke cost
829
+
830
+ ## 0.8.10: 2026-07-29
831
+
832
+ ### Added
833
+
834
+ - Scenarios carry stable SC-n IDs, coverage becomes mechanical
835
+ - Clickable prototype for UI features, prose is not reviewable
836
+
837
+ ## 0.8.9: 2026-07-29
838
+
839
+ ### Added
840
+
841
+ - Gate the repo against gitlinks and eval-workspace debris
842
+ - Arm just verify as a real pre-commit hook
843
+ - Gate the behavior-delta merge at Verdict: DONE
844
+
845
+ ### Documentation
846
+
847
+ - Add roadmap 018
848
+ - Update ROADMAP_018
849
+
850
+ ### Fixed
851
+
852
+ - Actually stage the env-leak fix from the previous commit
853
+
854
+ ## 0.8.8: 2026-07-28
855
+
856
+ ### Changed
857
+
858
+ - Default verifier and code-reviewer to sonnet
859
+
860
+ ### Documentation
861
+
862
+ - Guidance on picking the model per phase
863
+
864
+ ## 0.8.7: 2026-07-28
865
+
866
+ ### Added
867
+
868
+ - Derive the active spec from the branch, not from mtime
869
+
870
+ ### Changed
871
+
872
+ - Compute context cost in one awk pass
873
+ - Scope check-claims fixture mode to the fixture
874
+
875
+ ### Documentation
876
+
877
+ - Record the 0.8.7 correction pass
878
+
879
+ ## 0.8.6: 2026-07-28
880
+
881
+ ### Added
882
+
883
+ - Derive parallel-safety instead of trusting the marking
884
+ - Gate the per-command invoke-cost table
885
+
886
+ ### Changed
887
+
888
+ - Run the seven deterministic checks concurrently
889
+ - Load each phase when it is reached
890
+ - Pin tiers 31-33, and restore coverage the split dropped
891
+
892
+ ### Documentation
893
+
894
+ - Say where a cheaper model is the right call
895
+ - State that disjointness is now verified, and its limit
896
+ - ROADMAP_017, the framework's own efficiency measured and gated
897
+
898
+ ### Fixed
899
+
900
+ - Surface roster roles the handoff gate silently ignores
901
+
902
+ ## 0.8.5: 2026-07-27
903
+
904
+ ### Added
905
+
906
+ - Build the gate the docs described, wire it on every tool
907
+
908
+ ### Documentation
909
+
910
+ - Consodocs: update architecture flows, add project-type guide
911
+ - Update team-workflow.md with v0.8.4 handoff gate, members.md setup, and role examples
912
+ - List the learn skill in the core component row
913
+ - Clarify team workflow
914
+
915
+ ### Fixed
916
+
917
+ - Adapter DoD shims emit a real decision on both paths
918
+
919
+ ## 0.8.4: 2026-07-27
920
+
921
+ ### Added
922
+
923
+ - Scoring infrastructure and criteria for all 13 scenarios
924
+ - Detection script, conventions template, visual-check marker
925
+ - Quality-correlation, mine-conventions, transitive+timeline views
926
+ - Handoff gate, members template, security documentation
927
+ - Wire handoff gate into copilot and codex Stop hooks
928
+ - Add orphaned worktree with unmerged commits check
929
+ - Embed archetype conventions in plan (Tier 19)
930
+ - Add optional archetype sections to codebase-map template
931
+ - Add archetype-typical verification commands to conventions template
932
+
933
+ ### Changed
934
+
935
+ - Setup scenarios
936
+ - Regenerate handoff adapters after new skill
937
+ - Tick Tier 24 Known Volatility (already in write-spec.md:80)
938
+ - Tick 3 meta notes and ROADMAP_014 correlation item
939
+ - Tick worktree, claims, shims, failure modes, scale items
940
+ - Initialize scoring floor with all 13 scenarios
941
+ - Run 01-small-fix scenario end-to-end, result committed
942
+ - Tick OpenCode eval, scoring floor, failure modes, scale limits
943
+ - Validate Codex/Cursor/Copilot hook configs, close ROADMAP_013 gate-parity
944
+ - Discover hooks for the fail-open sweep instead of listing them
945
+ - Run the 1.0 audit, verdict is fail — 6 pass, 1 partial, 3 fail
946
+ - Record why T21 cannot be closed from inside this repo
947
+
948
+ ### Documentation
949
+
950
+ - Failure modes matrix, scale ceilings, 1.0 audit checklist
951
+ - Update scale-limits.md with real measurements
952
+ - Confirm every inferred marker, fix two wrong ones
953
+ - Induce 7 of 9 rows, two named checks that don't exist
954
+
955
+ ### Fixed
956
+
957
+ - Quality-correlation parsing, adopt-check, brownfield docs, conflict logging
958
+ - Test-integrity rubric counts assertion call sites, not lines
959
+ - Restore original line-matching with consolidation exception
960
+ - Mine-conventions rewrite, detect-archetype JSON, roadmap tick completed items
961
+ - Rename 06c criteria to match scenario name
962
+ - Stop the test-integrity and handoff gates blocking legitimate work
963
+ - Archetype detection never worked and was never invoked
964
+ - Findings-summary counted findings as completed specs
965
+ - Stop mandating a re-verify pass, fix the broken --timeline view
966
+ - Wire --reverse-engineer, correct four more roadmap ticks
967
+
968
+ ## 0.8.3: 2026-07-27
969
+
970
+ ### Added
971
+
972
+ - OpenCode as eval provider in `evals/run.sh` (`EVAL_PROVIDER=opencode`)
973
+ - `evals/with-opencode.sh` auto-installs DevOS plugin into project's `.opencode/` before running OpenCode scenarios
974
+ - Live OpenCode eval results: `evals/results/2026-07-27.md` (3 scenario rows from `opencode/ling-3.0-flash-free`)
975
+ - Per-tool hook I/O shim generation via `HOOKS` config in `scripts/tools.mjs`
976
+ - `generate-adapters.mjs` produces all 6 per-tool DoD and test-integrity shims from config
977
+ - `--check` drift detection verifies generated hook shims match committed shims
978
+
979
+ ### Changed
980
+
981
+ - Tier 12: Wired OpenCode into eval harness with `EVAL_PROVIDER=opencode` env var
982
+ - Tier 12: Updated support-matrix.md OpenCode CI row with new eval plumbing
983
+ - Tier 12: Updated known gaps section to reflect OpenCode eval harness progress
984
+ - Tier 13: Regenerated all 6 hook shims as canonical generator output
985
+ - Regenerated all hook shims from config (tool-specific headers replaced with generated ones)
986
+
987
+ ### Fixed
988
+
989
+ - Generated DoD/test-integrity shims are now generated (rule 21) instead of hand-maintained
990
+
991
+ ## 0.8.2: 2026-07-27
992
+
993
+ ### Added
994
+
995
+ - Onboard this repo — the first real devos/ folder
996
+ - Add time-pressure eval fixture (05d)
997
+ - Add reverse-engineer.sh, git-history-derived spec stubs for brownfield onboarding
998
+
999
+ ### Changed
1000
+
1001
+ - Complete run-sh-results-suffix, merge behavior delta
1002
+ - Fix evals/run.sh results-file truncation
1003
+ - Complete pressure-fixture-time-pressure, merge behavior delta
1004
+ - Time-pressure eval fixture + test-integrity hook fix
1005
+ - Complete reverse-engineer-spec-stubs, merge behavior delta
1006
+ - Reverse-engineer.sh spec stubs
1007
+
1008
+ ### Documentation
1009
+
1010
+ - Functional spec
1011
+ - Functional spec
1012
+ - Functional spec
1013
+ - Close ROADMAP_016 Tiers 10-11 with real dogfooding receipts
1014
+
1015
+ ### Fixed
1016
+
1017
+ - Stop evals/run.sh truncating a same-day results file
1018
+ - Test-integrity gate no longer misclassifies devos/specs/*.md as test files
1019
+ - Correct reverse-engineer-spec-stubs quality.md route to M (1 task group, not 3+)
1020
+
1021
+ ## 0.8.1: 2026-07-27
1022
+
1023
+ ### Added
1024
+
1025
+ - Give the 24 rules one home in RULES.md
1026
+ - Close the three ungated claim surfaces
1027
+ - Add devos-contract version marker to spec.md and conventions.md
1028
+
1029
+ ### Documentation
1030
+
1031
+ - Resolve all remaining [~] partial-verification items
1032
+
1033
+ ## 0.8.0: 2026-07-27
1034
+
1035
+ ### Added
1036
+
1037
+ - V0.8.0 Tier 1 — data-vs-instruction boundary
1038
+ - V0.8.0 Tier 2 — ambiguity markers in spec-write
1039
+ - V0.8.0 Tier 3 — convention pre-gate in implement
1040
+ - Add scoring rubric for gate-critical scenarios
1041
+
1042
+ ### Documentation
1043
+
1044
+ - Sharpen 015 Conditional — close stale gaps, name next move
1045
+ - Sharpen 015 Conditional — close stale gaps, name next move
1046
+ - Add roadmap 16
1047
+ - Improve roadmap and team roles
1048
+ - Improve roadmap and team roles
1049
+
1050
+ ## 0.7.3: 2026-07-26
1051
+
1052
+ ### Added
1053
+
1054
+ - Declare and derive Requires: dependency edges
1055
+
1056
+ ### Changed
1057
+
1058
+ - Re-deferrals for 014, update 015 conditional table
1059
+ - ROADMAP_014 complete — evals run via opencode/mimo-v2.5-free
1060
+ - ROADMAP_015 precondition updated — ROADMAP_014 complete
1061
+
1062
+ ### Documentation
1063
+
1064
+ - Document /devos:diff
1065
+ - Fix stale function name in evals.md
1066
+ - Add roadmap 015
1067
+ - Sharpen 015 Conditional — close stale gaps, name next move
1068
+
1069
+ ### Fixed
1070
+
1071
+ - Generate correctly path skills for each adapters
1072
+
1073
+ ## 0.7.2: 2026-07-25
1074
+
1075
+ ### Added
1076
+
1077
+ - Rebuild /devos:diff as a real, fixture-tested skill
1078
+
1079
+ ### Fixed
1080
+
1081
+ - Correct false-complete tiers, delete dead/broken scripts
1082
+
1083
+ ## 0.7.1: 2026-07-24
1084
+
1085
+ ### Fixed
1086
+
1087
+ - Resolve 5 regressions found in post-release audit
1088
+
1089
+ ## 0.7.0: 2026-07-24
1090
+
1091
+ ### Added
1092
+
1093
+ - Implement check-adapters drift report, spec quality signal, findings log, diff/timeline/doctor audit, and baseline evals
1094
+ - Add cached templates, test-integrity read-mask prototype, latest eval results
1095
+
1096
+ ### Changed
1097
+
1098
+ - Commit baseline-014.md results for scenarios 01-04
1099
+ - Parallel execution, cached templates, skip-done, configurable max-turns
1100
+
1101
+ ### Documentation
1102
+
1103
+ - Update docs, correct token cost
1104
+ - Add roadmap 014
1105
+ - Update parallel delegation notes and token-efficiency with baseline data
1106
+ - Mark all ROADMAP_014 items as completed
1107
+
1108
+ ## 0.6.6: 2026-07-24
1109
+
1110
+ ### Fixed
1111
+
1112
+ - Make the plugin self-contained so gates survive install
1113
+
1114
+ ## 0.6.5: 2026-07-23
1115
+
1116
+ ### Fixed
1117
+
1118
+ - Symlink devos-product/devos-docs skills on every adapter
1119
+
1120
+ ## 0.6.4: 2026-07-23
1121
+
1122
+ ### Added
1123
+
1124
+ - Extend generator reach to devos-docs
1125
+
1126
+ ## 0.6.3: 2026-07-23
1127
+
1128
+ ### Added
1129
+
1130
+ - Sharper specs before they lock, per-step checks
1131
+
1132
+ ### Changed
1133
+
1134
+ - V0.6.3 sharper specs, need-fit lens, per-step checks (ROADMAP_013)
1135
+
1136
+ ## 0.6.2: 2026-07-23
1137
+
1138
+ ### Added
1139
+
1140
+ - /devos:learn captures and routes durable lessons
1141
+
1142
+ ### Changed
1143
+
1144
+ - V0.6.2 learn loop (ROADMAP_013)
1145
+
1146
+ ## 0.6.1: 2026-07-23
1147
+
1148
+ ### Added
1149
+
1150
+ - Scoped project rules that attach at edit time
1151
+
1152
+ ### Changed
1153
+
1154
+ - V0.6.1 scoped rule layer (ROADMAP_013)
1155
+
1156
+ ## 0.6.0: 2026-07-23
1157
+
1158
+ ### Added
1159
+
1160
+ - Generate agent bodies and command wrappers from source
1161
+ - Deterministic test-integrity floor on Claude Code too
1162
+ - Extend generator reach to the devos-product plugin
1163
+
1164
+ ### Changed
1165
+
1166
+ - V0.6.0 generator, gate parity, devos-product reach (ROADMAP_013)
1167
+
1168
+ ### Documentation
1169
+
1170
+ - Update README
1171
+ - Add roadmap 013
1172
+ - Document the generator contract; reconcile roadmap with what shipped
1173
+
1174
+ ## 0.5.1: 2026-07-22
1175
+
1176
+ ### Added
1177
+
1178
+ - Author gate-under-pressure fixtures, authority and sunk-cost
1179
+
1180
+ ### Changed
1181
+
1182
+ - Baseline run 2026-07-19, single trial, no cap hit
1183
+ - Full suite run with pressure fixtures, no cap hit
1184
+ - Use devos: command syntax to match Claude Code
1185
+
1186
+ ### Documentation
1187
+
1188
+ - Untracked local config does not travel to a worktree, say so
1189
+ - State allowed values per Branching & workspace line, concrete setup examples
1190
+ - Restructure Branching & workspace section, value table and filled example
1191
+ - Configuration reference, every setting with values, defaults, and readers
1192
+ - Catch glossary, implement page, and resuming-work up to branch/workspace behavior
1193
+ - Explicit manual-cleanup section, exact commands, disk cost while a worktree lives
1194
+ - Move 012 deferred items to explicitly rejected, no active requester
1195
+ - Check baseline/parallel eval boxes, findings and fixes documented
1196
+ - Check gate-under-pressure box, mixed findings documented
1197
+
1198
+ ### Fixed
1199
+
1200
+ - Number Phase B pre-steps so resume cannot skip the branch check
1201
+ - Make ∥ dispatch a MUST, not an optional-reading bullet
1202
+ - Generalize pre-steps beyond resume, any urgency framing
1203
+ - Make claude-context more precise
1204
+ - Drop bogus opencode.jsonc plugin registration
1205
+ - Correct hooks.json to nested schema
1206
+
1207
+ ## 0.5.0: 2026-07-18
1208
+
1209
+ ### Added
1210
+
1211
+ - Spec worktrees, L-route isolation by default, declared setup
1212
+ - Flag merged-but-present spec worktrees, gate-tested
1213
+
1214
+ ### Documentation
1215
+
1216
+ - Parallel-work situation page, workspace boundary in architecture and contract
1217
+ - Check 012 tiers 1-2, shipped
1218
+
1219
+ ## 0.4.2: 2026-07-18
1220
+
1221
+ ### Added
1222
+
1223
+ - Spec branch before first edit, default branch is the human's
1224
+
1225
+ ### Changed
1226
+
1227
+ - Add roadmap_012
1228
+
1229
+ ### Documentation
1230
+
1231
+ - Add flow principle
1232
+ - Add ROADMAP_012, a place to work
1233
+ - Branch-per-spec model, recovery entry for commits on the default branch
1234
+
1235
+ ## 0.4.1: 2026-07-17
1236
+
1237
+ ### Added
1238
+
1239
+ - Gate the documented 500-line SKILL.md budget
1240
+ - Pin eval model, add trial count
1241
+ - Add evals recipe, point docs at it
1242
+
1243
+ ### Documentation
1244
+
1245
+ - Explain why conventions.md isn't a best-practices library
1246
+
1247
+ ## 0.4.0: 2026-07-17
1248
+
1249
+ ### Added
1250
+
1251
+ - Behavior delta section in functional specs
1252
+ - Behavior files merged at completion, specs archive stage
1253
+
1254
+ ### Documentation
1255
+
1256
+ - Check 011 tiers 5-7, shipped
1257
+
1258
+ ## 0.3.5: 2026-07-17
1259
+
1260
+ ### Added
1261
+
1262
+ - Scenario discipline, normative wording, removal ritual
1263
+ - Validate-spec checks spec folders mechanically, uniform diagnostic envelope
1264
+ - First-party review skill, calibrated severity, intent-gap work preservation
1265
+
1266
+ ### Documentation
1267
+
1268
+ - ROADMAP_011, the behavior layer
1269
+ - Troubleshooting, reviewing-changes, team-workflow, glossary, agent-contract, worked example, greenfield note
1270
+ - Check 011 tiers 1-4, shipped
1271
+
1272
+ ### Fixed
1273
+
1274
+ - Sync verifier scenario-coverage line into adapter copies
1275
+
1276
+ ## 0.3.4: 2026-07-16
1277
+
1278
+ ### Added
1279
+
1280
+ - Make entry-point skills user-invoked-only again
1281
+ - Gate token claims and command-prefix integrity in just verify
1282
+ - Map freshness signal in build/implement, size budget on the map
1283
+ - Doctor checks adoption health, not just installation
1284
+
1285
+ ### Documentation
1286
+
1287
+ - Update always-loaded claim to measured ~250 tokens
1288
+ - Script-ownership table and ROADMAP_010 (claims become gates)
1289
+
1290
+ ## 0.3.3: 2026-07-16
1291
+
1292
+ ### Documentation
1293
+
1294
+ - Add FAQ answering common objections (pasted ticket, single skill, monorepo, cost)
1295
+ - Pull comparison-table content into FAQ, generalized without naming external tools
1296
+ - Cleanup docs
1297
+ - Add architecture and explain about token usage principle
1298
+ - Fix command prefixes for devos-docs and devos-product modules
1299
+ - Name module entry commands with full prefixes in getting-started
1300
+ - Clarify agent-facing vs human-facing module output
1301
+
1302
+ ## 0.3.2: 2026-07-16
1303
+
1304
+ ### Added
1305
+
1306
+ - Routing collision and fixture-coverage gates, skill-authoring rules (009 tier 4)
1307
+ - Doctor command, version drift, gate scripts, adapter link health (009 tier 5)
1308
+
1309
+ ### Documentation
1310
+
1311
+ - Project-type guidance, wrapper-repo recipe, roadmap forward pointer (009 tier 6)
1312
+
1313
+ ## 0.3.1: 2026-07-16
1314
+
1315
+ ### Added
1316
+
1317
+ - Implementation discipline, clean baseline, named staging, decision self-challenge (009 tier 1)
1318
+ - Verifier depth, multi-line stub detection, blast-radius check order (009 tier 2)
1319
+ - Project-declared verification, proof commands in conventions, human-checked criteria (009 tier 3)
1320
+
1321
+ ### Changed
1322
+
1323
+ - Add roadmap 009
1324
+
1325
+ ## 0.3.0: 2026-07-16
1326
+
1327
+ ### Added
1328
+
1329
+ - Sharpen routing and approval gates (008 tier 1)
1330
+ - Harden the spec capsule (008 tier 2)
1331
+ - Verification protocol, asymmetry, triage, deferred ledger (008 tier 3)
1332
+ - Discipline rationalization tables and anti-summary lint (008 tier 4)
1333
+ - Token-free self-enforcement, gate tests, routing check, release tooling (008 tier 5)
1334
+
1335
+ ### Changed
1336
+
1337
+ - Add comparision tools
1338
+ - Add roadmap 008
1339
+
1340
+ ### Documentation
1341
+
1342
+ - Conventions baseline and 008 doc sync (008 tier 6)
1343
+
1344
+ ## 0.2.0: Parallel execution (token-neutral concurrency)
1345
+
1346
+ - Two admitted shapes of parallelism, both audited to cost zero extra tokens or fewer: concurrent commands inside one agent, and partitioned subagent fan-out where each branch's file set is disjoint from every other branch's.
1347
+ - `/devos:cross-review` now runs its Gemini and Codex reviews concurrently instead of sequentially, identical prompt, identical tokens, wall-clock drops from the sum of both to roughly the slower of the two.
1348
+ - `/devos:cross-review` degrades honestly when neither external CLI is installed: it now says so plainly and points to the harness's own review feature, instead of demanding an install. It never falls back to the harness's own model reviewing its own output, that would rebuild a review the harness already ships and fake the model diversity that is this skill's only reason to exist.
1349
+ - The verifier's quality lens now runs read-only project checks (lint, type-check) concurrently, then build/test checks (which write artifacts or share state) as a batch afterward.
1350
+ - `/devos:implement`'s plan template marks task groups with pairwise-disjoint file sets as parallel-safe (`Group A ∥ Group B, then Group C`), visible at the existing plan-approval gate, no new gate, no new state file. At execution, parallel-safe groups dispatch as concurrent isolated subagents, each reading only the spec capsule and its own group; a merge conflict demotes that group to sequential re-execution rather than attempting a resolution. Never marked below 3 task groups, S/M-sized work stays untouched.
1351
+ - `/devos:onboard --full`'s per-subsystem exploration (already partially in place) is now explicit about applying only to full scans, never to a small incremental re-scan.
1352
+ - `/devos:stats` reports per-branch token cost when a parallel route ran, so the "no more tokens than sequential" claim is checkable per project.
1353
+ - New eval scenarios `06-parallel-vs-sequential` and `06b-conflict-probe` (the gate-breaking test: task groups that share a file must refuse to parallelize) are committed and ready, not yet run against a live model, so `docs/token-efficiency.md` states "not yet measured" rather than inventing a number. Every "faster"/"no more tokens" claim in this release sits next to a real measurement or an honest admission that one doesn't exist yet.
1354
+ - `docs/support-matrix.md` gains a "Parallel delegation" row: native and tuned on Claude Code this release; Codex CLI and OpenCode both have some form of concurrent-subagent capability that isn't adapted yet; Cursor CLI and Copilot CLI have no documented mechanism, so the same prose reads sequentially there by construction.
1355
+ - Retroactively tagged `v0.0.6`, `v0.1.0`, and `v0.1.1`, those releases were versioned in `plugin.json` and documented in this changelog when they shipped, but the git tags themselves were missed at the time.
1356
+
1357
+ ## 0.1.1: Support-matrix convergence
1358
+
1359
+ - 5-column `docs/support-matrix.md` (Claude Code, Cursor CLI, Copilot CLI, Codex CLI, OpenCode); `docs/install/README.md` and the README "Works with" table updated to list all four adapters.
1360
+ - `evals/check-adapters.sh` now covers the two new adapters' duplicated agent bodies (`.toml` for Codex, `.md` for OpenCode), including a TOML-aware body extractor.
1361
+ - All three plugins bumped to `0.1.1`.
1362
+
1363
+ ## 0.1.0: OpenCode adapter (real code, not just symlinks)
1364
+
1365
+ - DevOS now runs on OpenCode, installed via `.opencode/scripts/install-opencode.sh`, symlink-based, `git pull` updates live. Unlike Cursor/Copilot/Codex, this adapter needed real code: OpenCode's skill/command/agent/hook mechanics don't map to translated frontmatter alone (ROADMAP_004 rule 11, one real pattern gap gets a real adapter).
1366
+ - The 11 orchestrators get command wrappers (`.opencode/commands/devos-*.md`, `$ARGUMENTS` supported); the 3 discipline skills and 11 orchestrator skills are symlinked directly (OpenCode's own `skill` tool discovers them); `verifier` and `code-reviewer` are translated to OpenCode's agent format (`mode: subagent`).
1367
+ - Test-integrity is mechanically enforced via a `tool.execute.before` plugin hook (`.opencode/plugins/devos.js`, 90 lines) that denies by throwing.
1368
+ - Definition-of-Done gets a new, honestly-named category, **enforced-via-correction**: OpenCode has no turn-end hook with transcript access, so the plugin reacts to the `session.idle` event, reads the transcript via the SDK client, and injects a correction message rather than blocking outright.
1369
+ - This adapter's hook mechanics were checked against a real, locally installed `opencode` (v1.17.18) and its `@opencode-ai/plugin`/`@opencode-ai/sdk` type definitions, not doc-only. The plugin's own hook logic was unit-tested directly (5 synthetic cases); a full LLM-driven session was not run, to avoid spending API credits without asking.
1370
+ - `docs/install/opencode.md` and an OpenCode column in `docs/support-matrix.md`.
1371
+
1372
+ ## 0.0.6: Codex CLI adapter
1373
+
1374
+ - DevOS now runs on Codex CLI, installed via `devos-codex/scripts/install-codex.sh`, symlink-based, `git pull` updates live.
1375
+ - Re-verifying this roadmap's capability snapshot at implementation time found it stale: Codex CLI hooks and subagents have both shipped since it was written (hooks GA 2026, subagents GA March 2026), upgrading this adapter from the originally-planned "gates degrade to prose" to full gate parity with Copilot CLI.
1376
+ - Both gates are enforced: Definition-of-Done via a `Stop` hook (`transcript_path` in the payload, reusing the same `scripts/gates/check-dod-core.sh` logic as every other adapter), test-integrity via a `PreToolUse` hook denying weakened test edits.
1377
+ - Orchestrators need no command-wrapper layer, Codex reads `SKILL.md` natively via its own skills mechanism, same shape as the Claude Code plugin itself. `verifier` and `code-reviewer` are translated to Codex's TOML custom-agent format (no forced invocation syntax, natural-language delegation only).
1378
+ - No Codex CLI binary was available to test hands-on; every claim is doc-verified and corroborated across multiple independent sources (official docs plus `openai/codex` GitHub issues/PRs), stated plainly in `docs/support-matrix.md`.
1379
+ - `docs/install/codex.md` and a Codex CLI column in `docs/support-matrix.md`.
1380
+
1381
+ ## 0.0.5: Doc fixes following the 0.0.4 release
1382
+
1383
+ - Every existing doc page that showed `/devos:` syntax now says plainly it's Claude Code's, and links to the Cursor/Copilot equivalent and the support matrix, the 0.0.4 release had only added new pages without updating the ones that already existed.
1384
+ - Corrected a claim that had been wrong since 0.0.1: `git pull` alone does **not** update a `claude plugin install`-ed copy of DevOS. `claude plugin install` copies a version-pinned snapshot into `~/.claude/plugins/cache/`; confirmed directly that the cache stays on the old version after a `git pull` until `claude plugin update devos@devos` (+ restart) is run. `README.md`, `docs/concepts.md`, and `docs/install/` now state the real mechanism. Cursor CLI and Copilot CLI are unaffected, their adapters install via symlink, which genuinely is `git pull`-live.
1385
+
1386
+ ## 0.0.4: Cursor CLI and Copilot CLI support
1387
+
1388
+ - DevOS now runs on Cursor CLI and GitHub Copilot CLI, installed via `.cursor/scripts/install-cursor.sh` and `.github/scripts/install-copilot.sh`, symlink-based, `git pull` updates live, same posture as the Claude Code install.
1389
+ - Both adapters get the full orchestrator set (`build`, `spec`, `spec-write`, `implement`, `specs`, `verify-gap`, `cross-review`, `onboard`, `domain-model`, `claude-context`, `stats`) and the two subagents (`verifier`, `code-reviewer`).
1390
+ - Test-integrity is mechanically enforced on both new tools (deterministic weakened-assertion detection, shared logic in `scripts/gates/test-integrity-rubric.sh`).
1391
+ - The Definition-of-Done gate is mechanically enforced on Copilot CLI (same guarantee as Claude Code) and is a documented prose-only discipline on Cursor CLI, because Cursor's stop hook has no transcript access to check against, recorded plainly in `docs/support-matrix.md` rather than shipping a hook that does nothing.
1392
+ - New `docs/support-matrix.md` and `docs/install/` (one page per tool), every capability cell states enforced, prose, or n/a with the reasoning behind it.
1393
+ - New `evals/check-adapters.sh` catches drift between the canonical skill/agent bodies and their per-tool copies.
1394
+ - Every existing doc page that shows `/devos:` syntax (`docs/README.md`, `docs/build.md`, all 9 `docs/commands/*.md` pages, and the `docs/situations/*.md` pages) now says plainly that it's Claude Code's syntax and links to the equivalent Cursor/Copilot invocation and the support matrix, the first pass at this release had only added new pages without updating the existing ones, which still read as Claude-only.
1395
+
1396
+ ## 0.0.3: Harness-neutral core
1397
+
1398
+ - Implementation, spec, domain-model, spec-write, verify-gap, and cross-review skills now describe plan approval, structured Q&A, task tracking, and subagent delegation as behaviors instead of naming Claude-specific tool calls, the same prose now reads correctly on any harness that supports the same primitives.
1399
+ - The project-context workflow writes the same snippet into `AGENTS.md`, so the artifact wiring survives across supported harnesses.
1400
+
1401
+ ## 0.0.2: First tagged release
1402
+
1403
+ - First git tag (`v0.0.1`) for the framework as it stood before this release.
1404
+ - Added a committed eval scenario proving the test-integrity gate holds under a realistic temptation (a failing test with a real source bug, where weakening the assertion is the easy way out).
1405
+ - `CHANGELOG.md` introduced.
1406
+
1407
+ ## 0.0.1: Initial version
1408
+
1409
+ Everything up to this point, never published before now:
1410
+
1411
+ - **Adaptive entry point** (`/devos:build`): sizes any request (S/M/L/R) and routes to the matching amount of process, no ceremony for a small fix, a full spec pipeline only when the task actually needs one.
1412
+ - **Spec pipeline**: `/devos:spec` (interactive Q&A), `/devos:spec-write` (synthesis into a validated functional spec), `/devos:implement` (plan in native plan mode, one approval gate, implement, verify).
1413
+ - **spec.md as a context capsule**: the implementing session reads one file, conventions, domain vocabulary, and prior decisions are embedded in the plan, not re-fetched.
1414
+ - **Enforced gates**: a Definition-of-Done stop hook blocks a false "done" without a verification report; a test-integrity hook denies edits that weaken tests to make them pass.
1415
+ - **Composable verification**: one verifier with two lenses (completeness + quality), findings passed through an adversarial refute pass before being reported; `/devos:verify-gap` for a completeness-only check; `/devos:cross-review` for a second opinion from other AI models on a git diff.
1416
+ - **Project knowledge artifacts**: `conventions.md`, `domain-model.md`, `codebase-map.md`, `decisions.md`, versioned, incrementally updated, shared vocabulary across every skill.
1417
+ - **`/devos:onboard`**: maps an existing codebase into the knowledge artifacts above; incremental re-scan on repeat runs.
1418
+ - **`/devos:stats`**: measures the framework's own always-loaded token cost and per-skill load cost.
1419
+ - **Headless support**: spec-write and verify-gap run non-interactively for CI.
1420
+ - **Optional modules**, installed separately, zero cost unless used: `devos-docs` (documentation system) and `devos-product` (mission, roadmap, backlog, reverse-engineering existing codebases into product docs).
1421
+ - **Distribution**: a local Claude Code plugin marketplace pointed at a git clone, nothing published. **Correction (0.0.4):** "`git pull` updates live" as originally claimed here turned out to be wrong for the installed plugin content itself, confirmed directly, `claude plugin install` copies a version-pinned snapshot; `claude plugin update devos@devos` plus a restart is required. See `docs/install/claude-code.md#update`.