gcf-aif-ext 0.32.0 → 0.36.0
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.
- package/README.md +45 -6
- package/commands/gcf.js +344 -82
- package/extension.json +1 -1
- package/injections/plan-context.md +3 -0
- package/lib/agent-permissions.js +28 -4
- package/lib/audit-core.js +72 -127
- package/lib/board.js +22 -10
- package/lib/changes.js +7 -1
- package/lib/commit-files.js +1 -0
- package/lib/commit-move.js +56 -10
- package/lib/config-set.js +52 -14
- package/lib/context.js +29 -16
- package/lib/d2-states.js +208 -0
- package/lib/design-parts.js +37 -7
- package/lib/domain.js +144 -11
- package/lib/gcf-paths.js +13 -2
- package/lib/git.js +62 -17
- package/lib/inbox-categories.js +3 -0
- package/lib/inbox.js +69 -4
- package/lib/lessons.js +133 -4
- package/lib/main-lock.js +195 -3
- package/lib/plan-check.js +339 -37
- package/lib/play.js +21 -20
- package/lib/review.js +111 -2
- package/lib/rules.js +428 -0
- package/lib/session-mark.js +2 -1
- package/lib/systems-doc.js +60 -13
- package/lib/validate-item.js +9 -0
- package/lib/work-ops.js +109 -50
- package/package.json +1 -1
- package/profiles/plan.md +1 -1
- package/schemas/item.schema.json +4 -0
- package/skills/gcf-audit/SKILL.md +10 -5
- package/skills/gcf-card/SKILL.md +12 -10
- package/skills/gcf-chat/SKILL.md +14 -10
- package/skills/gcf-triage/SKILL.md +19 -1
- package/skills/gcf-verdict/SKILL.md +7 -5
- package/templates/agents-block.md +23 -2
- package/templates/config.yml +10 -6
package/README.md
CHANGED
|
@@ -50,6 +50,44 @@ Updating preserves `gcf/`, `cnlp.config.json` and the pre-commit hook. Run `ai-f
|
|
|
50
50
|
|
|
51
51
|
On Windows, run `extension add` and `extension update` from PowerShell or cmd, not Git Bash: the installer unpacks with `tar`, and Git Bash's GNU tar reads `C:\…` as a remote host (`Cannot connect to C: resolve failed`). The Windows `tar.exe` handles it.
|
|
52
52
|
|
|
53
|
+
### 0.36.0
|
|
54
|
+
|
|
55
|
+
Update together with gcf-harness-aif-ext 0.35.0 and gcf-board 0.36.0 ([Plan-Review-Rework-Plan](https://github.com/VirtualMaestro/GCF/blob/main/docs/Plan-Review-Rework-Plan.md), Board-PRD decision 67). Then run `ai-factory gcf init`: the `AGENTS.md` block changed.
|
|
56
|
+
|
|
57
|
+
- **A plan review finds, the plan's author judges.** The reviewer only finds; the session that wrote the plan checks each finding against the code, gives it a verdict (`accepted`, `rejected`, `downgraded`) and fixes the plan. `gcf plan pass <id> --findings <file.json> --by …` writes the findings with their verdicts to the plan's `## Findings` (`- [c<C>.]p<K>.<N> <blocking|polish> <verdict> <d#/t#>: <summary>[; <reason>]`) and counts the pass line itself; `--blocking N --polish M` stays for an older harness and is refused in a cycle that already has findings. A blocking finding with no `scenario` is recorded as polish. The review converges when the last pass has no blocking finding with any verdict, so a rejected or downgraded one is followed by a check.
|
|
58
|
+
- **Full review, then checks.** `gcf review row <id> --kind plan --pass <k> [--diff] --json` keeps the plan as that reviewer saw it in `.gcf-session/plans/<id>.md` and answers `mode` (`full`, or `check` of the fixes) and `cycle`, with `--diff` the `diff.file` of the plan's changes since the earlier pass; it asks only the next pass and refuses a stalled plan. `gcf plan check --json` carries `findings`, and prints a `format:` line for a plan over about 150 lines or 20 tasks.
|
|
59
|
+
- **A model and effort per run phase.** `plan_model`, `plan_effort`, `implement_model`, `implement_effort` on a card (`gcf item agent <id> --phase plan|implement --model <m> --effort <e>`) and `agent.plan.model|effort`, `agent.implement.model|effort` in `gcf/config.yml` (`gcf config set agent.plan.effort high`, keys of three levels); `gcf board --json` carries them and `agent.phase_defaults` (GCF's own: `plan.effort: high`).
|
|
60
|
+
- **Review defaults:** `review.plan` is `codex:astra:high`, `claude:fable:medium`, `claude:fable:medium`; lists already in `gcf/config.yml` stay as they are.
|
|
61
|
+
|
|
62
|
+
### 0.35.0
|
|
63
|
+
|
|
64
|
+
Update together with gcf-board 0.35.0 (`gcf board --json` is `schema: 7`; each refuses the other's old schema) and gcf-harness-aif-ext 0.34.0 ([Lessons-Review-Plan](https://github.com/VirtualMaestro/GCF/blob/main/docs/Lessons-Review-Plan.md), Board-PRD decision 66). Then run `ai-factory gcf init`: the `AGENTS.md` block changed (groups of entries, the project's rules, lesson reports, the base-improving tech note).
|
|
65
|
+
|
|
66
|
+
- **A group of inbox entries of one cause is one move and one commit.** `item new --from in-a,in-b` and `item amend --from in-a,in-b` put a bullet of each, its id first, in the note (`item new` takes the first entry's title unless `--title`; past 4000 characters it refuses: shorten or split the group, unless `--note`/`--note-file` gives the note, which then takes the bullets' place; their pictures are named after their entry's id); `inbox drop in-a in-b` and `inbox later in-a in-b` check every entry before they write any, and one refused leaves all as they are. `inbox drop --reason done|duplicate|noise|reported|amended [--of <id|hash>]` puts the reason in the commit's subject and a `GCF-Reason:` trailer; an erase of entries never committed still commits, so its trailers stand. `--proposal <thread>/<id>` on `item new`, `item amend`, `inbox drop`, `inbox later`, `rules add` and `lessons report` names the board's proposal in a `GCF-Proposal:` trailer; the domain flags those moves `proposal_trailer`, and a board recovers a proposal made only by that commit.
|
|
67
|
+
- **The project's own rules.** `gcf rules [--json]` reads `AGENTS.md` and `CLAUDE.md` (`{files: [{file, text, base, bytes, block, imports_agents, dirty}], own_rules}`, `own_rules` the lines under `## Project rules` of `AGENTS.md`). `rules add --text <line> [--from <in-ids>] [--base <hash>] (--dry-run [--json] | --commit)` puts one line, up to 300 characters, under `## Project rules`, outside GCF's block, of `AGENTS.md` and of a `CLAUDE.md` that does not import it; refused: a repeat, several lines, a marker, a leading `@` or `#`, an `@path` anywhere, a code fence; `--base` refuses files changed since the dry run that gave it; `gcf init` keeps the line. `rules save --file AGENTS.md|CLAUDE.md --base <hash> --text-file - (--dry-run | --commit)` saves a whole file in its own line endings, refused for a stale base, a changed GCF block or a marker pair too many or too few. A file with edits not committed (a `gcf init` refresh) goes into the commit whole; `--dry-run` shows its diff from HEAD.
|
|
68
|
+
- **Lesson reports.** `lessons report --from <in-ids> --to framework|template (--dry-run [--json] | --commit)` writes a file a lesson to `<GCF_BOARD_HOME or ~/.gcf>/lessons/<to>/<project>-<in-id>.md` (the project, the card, the kind, when it was recorded, the lesson's text with its `where` as the last line, the core, harness and board versions — the board's from `GCF_BOARD_VERSION` — and the date; a file already there, of a lesson still in the inbox, is a leftover of an attempt stopped before its commit and is written over) and erases the entries (`reported`) in one commit; an entry not of the category `lesson` is refused.
|
|
69
|
+
- `rules add`, `rules save` and `lessons report` are the human's moves: a run and a thread are refused, the deny group `human-moves` holds them; a chat agent proposes `rules add` and `lessons report` (`propose_rules_add`, `propose_lessons_report`).
|
|
70
|
+
- **The instructions mark.** `rules diff --from <hash> --to <hash> [--names] [--json]` (hex only) gives what the commits `from..to` changed in `AGENTS.md`, `CLAUDE.md`, `.claude/` and `.ai-factory/` without its plans and archive (`INSTRUCTION_PATHS`): no merges, first parent, no move of the operator's. `gcf changes --json` marks such a file `instructions: true`.
|
|
71
|
+
- The domain: the thread start `lessons-review` (skill gcf-triage; its agent may not edit or name in a shell command the instruction files, YOLO or not; `entries: lesson`, `next`), the moves `rules-add`, `lessons-report`, `rules-save`, `inbox-drop` and `inbox-later` on `{ids}` with the erase's `reason` and `of`, `from` as an `ids` field (`inbox: true`), and `sections.lessons {review, rule, report}`. gcf-triage teaches the review by groups; gcf-chat the new proposals.
|
|
72
|
+
|
|
73
|
+
### 0.34.0
|
|
74
|
+
|
|
75
|
+
Update together with gcf-board 0.34.0 and gcf-harness-aif-ext 0.33.0 ([Until-Records-Plan](https://github.com/VirtualMaestro/GCF/blob/main/docs/Until-Records-Plan.md)). Then run `ai-factory gcf init`: the `AGENTS.md` block changed (a session starts with `ai-factory gcf audit` and states its first line; a fired D2 goes to the human, who answers on the record itself). If the audit then reports inbox entries from before the categories, run `ai-factory gcf inbox migrate` once.
|
|
76
|
+
|
|
77
|
+
- **A fired assumption reaches the operator.** A D2 whose `until:` fired and has no `answer:` is a stale until record the audit's first line names (`stale until-records: q-019 retune(input) — ask the human`); an answered one is `answered`, never stale. `gcf board --json` gives such a D2 `category: decision` with `fired: true`, and a D2 of `manual(...)` or no trigger left unanswered past `question_stale_after_commits` (20) `category: decision` with `aged: true` and `commits`: the board shows both among the decisions. The human answers on the record (`gcf question answer q-NNN`, or Overrule on the board), never as a new D1; the agent then writes the answer into each slot citing `assumed: q-NNN` and drops the record, except while an unplayed item cites it.
|
|
78
|
+
- **`retune(<sys>)` counts only the tunes after the record.** It fires on a tune verdict of the subsystem after the commit that wrote the D2 (or the annotation's own line); a tune before it no longer makes a new record stale at birth, an uncommitted record is `open`, and a record outside git keeps the old rule: it fires once an item of the subsystem was tuned, or on any tune in the log.
|
|
79
|
+
- `stale_questions` keeps what the human still owes: unanswered D1, and unanswered D2 of `manual(...)` or no trigger past the limit; answered records and live `verdict()`/`retune()` ones are out.
|
|
80
|
+
- **GCF's own placeholders close by themselves.** `[until: manual(purpose of <s>)]` and `[until: manual(first verdict touching <s>)]` go when content arrives (`verdict keep`, `done --constraint`); writing a subsystem's purpose (`design save --kind system`, an import, a design commit) puts it in place of the marker in an existing `systems/<s>.md` in the same commit. Old markers in a project go at the next keep or design save of that subsystem. A marker left is listed in `audit --json` with `placeholder: true`, never in the first line or on the board. A human's own `manual()` stays.
|
|
81
|
+
- The readset takes a D2 an item cites as `assumed:` whatever its `sys:`, and lists `d2_fired`; plain `gcf context` prints each D2 record whole, `[fired: <trigger>]` on a fired one.
|
|
82
|
+
|
|
83
|
+
### 0.33.0
|
|
84
|
+
|
|
85
|
+
Update together with gcf-board 0.33.0 and gcf-harness-aif-ext 0.32.0 ([Main-Run-Moves-Plan](https://github.com/VirtualMaestro/GCF/blob/main/docs/Main-Run-Moves-Plan.md)). Then run `ai-factory gcf init`: the `AGENTS.md` block changed.
|
|
86
|
+
|
|
87
|
+
- **An operator's move waits behind a card's main run instead of a refusal.** A move from a terminal (stdin is a tty) or from a board that passes `GCF_MOVE_TAG`, refused as `dirty` while a card's main run holds the checkout, waits without the main lock: it says `<move> waits: the run of <card> holds gcf/ edits (…)` and a `gcf-waiting: {"behind","paths"}` line on stderr, keeps `.gcf-session/waiting/<pid>.json`, looks once a second, and goes again from its first read once the refusal's paths are clean. Nothing is written while it waits. When the run ends with the paths still dirty (your own `gcf/` edits), it gets today's refusal. Ctrl-C, or the file `.gcf-session/waiting/<pid>.cancel` (`GCF_MOVE_CANCEL` names another), cancels it: nothing recorded.
|
|
88
|
+
- Sessions, threads, runs, a script's move without a tty or `GCF_MOVE_TAG`, and builds, push and play do not wait: refused as before. A card's claim yields up to 5 s to a waiting move about to go.
|
|
89
|
+
- With an older board (no `GCF_MOVE_TAG`) a move from the board is refused as before.
|
|
90
|
+
|
|
53
91
|
### 0.32.0
|
|
54
92
|
|
|
55
93
|
Update together with gcf-board 0.32.0 (`gcf board --json` is `schema: 6`; each refuses the other's old schema) and gcf-harness-aif-ext 0.31.0 ([Fallback-Fidelity-Plan](https://github.com/VirtualMaestro/GCF/blob/main/docs/Fallback-Fidelity-Plan.md)). Then run `ai-factory gcf init`: the `AGENTS.md` block changed.
|
|
@@ -76,7 +114,7 @@ Update together with gcf-board 0.32.0 (`gcf board --json` is `schema: 6`; each r
|
|
|
76
114
|
|
|
77
115
|
### 0.23.0
|
|
78
116
|
|
|
79
|
-
- `gcf item amend <id>` fixes a live card's text from every stored status but `unplayed`, the status unchanged: `--text` (`--text-file`) replaces the blocks it names, a key alone takes one out; `--add` adds to `notes:`; `--title`; `--from <in-
|
|
117
|
+
- `gcf item amend <id>` fixes a live card's text from every stored status but `unplayed`, the status unchanged: `--text` (`--text-file`) replaces the blocks it names, a key alone takes one out; `--add` adds to `notes:`; `--title`; `--from <in-ids>` takes inbox entries, comma-separated, in the same commit, its pictures one by one. `provenance:` and `tuning:` cannot be named; a unit's block must be in its profile. `--dry-run --json` writes nothing and gives `{base, title, text, changes}`; `--base <hash>` refuses a card changed since (`gcf-refusal: {"code":"stale"}`). The operator's: a run and a thread are refused, the deny group `human-moves` holds it, and a chat agent proposes it (`propose_amend`). Refused during this card's run or session (`{"code":"running"}`). Findings on the new text come from the pre-commit gate, with its rollback. The domain's `preview` and `hint` tell a board.
|
|
80
118
|
- A resumed worktree takes main's card over what a stopped run wrote into its copy, saved or not; other files' conflicts read as before. A `work/<id>` branch whose folder was deleted syncs with main too.
|
|
81
119
|
- `item new --from <in> --note <text>`: the note goes over the entry's text, not beside it, so a board's form does not double it. The entry's links follow its pictures.
|
|
82
120
|
- The domain's `inbox-add` has the `item` field: an agent's proposal may name its card.
|
|
@@ -182,7 +220,7 @@ Autonomy (docs/Autonomy-Plan.md): the agent works by itself, the operator plays
|
|
|
182
220
|
|
|
183
221
|
Update together with gcf-board 0.15.0 (`gcf board --json` is `schema: 4`; each refuses the other's old schema) and the gcf-harness-aif-ext released with it. Then run `ai-factory gcf init` (the AGENTS.md block learns the categories and `item new --title`) and, once, `ai-factory gcf inbox migrate`.
|
|
184
222
|
|
|
185
|
-
- **The inbox by categories** (Board-PRD decision 32). An entry's frontmatter is `category`, `detail`, `origin`, `later`, `item`, `at`; `source` and `tags` are gone. The category is one of a fixed vocabulary, the core's base and the domain's layer merged (`ai-factory gcf inbox categories [--json]` prints it with who may file each); the project's layer is reserved. `inbox add [--category <c>] [--detail <word>] [--origin you|agent|automation] [--item <id>]`: the operator's entry with no category is a `note`, an agent must name one; the origin defaults to who makes the move (an agent's run or session is `agent` and may claim no other). `inbox category <in-id> [--category] [--detail]` replaces `inbox tag`: the operator moves an entry to any category but `decision`, `assumption` and `unsorted`, an agent only to one it may file; the origin never changes. `inbox later <in-
|
|
223
|
+
- **The inbox by categories** (Board-PRD decision 32). An entry's frontmatter is `category`, `detail`, `origin`, `later`, `item`, `at`; `source` and `tags` are gone. The category is one of a fixed vocabulary, the core's base and the domain's layer merged (`ai-factory gcf inbox categories [--json]` prints it with who may file each); the project's layer is reserved. `inbox add [--category <c>] [--detail <word>] [--origin you|agent|automation] [--item <id>]`: the operator's entry with no category is a `note`, an agent must name one; the origin defaults to who makes the move (an agent's run or session is `agent` and may claim no other). `inbox category <in-id> [--category] [--detail]` replaces `inbox tag`: the operator moves an entry to any category but `decision`, `assumption` and `unsorted`, an agent only to one it may file; the origin never changes. `inbox later <in-ids...> [--off]` puts entries away to Later and back, one commit; `decision`, `failure` and `assumption` never go there. The domain's moves are `inbox-add`, `inbox-category`, `inbox-later`, `inbox-back`, `inbox-drop`; an agent may propose the first three and `inbox-drop`.
|
|
186
224
|
- `inbox migrate [--json]` writes the entries from before the categories in the new format, in one commit `gcf(inbox): N entries to categories`, and lists what did not fit: `playtest` stays, `ci` → `failure`, `agent` → `unsorted`, `operator` → `note`, a lesson → `lesson` with its kind as the detail and its `item`, the tag `later` → the flag, the first other tag → the detail, the rest dropped. A second run changes nothing. Until it runs, the core reads an old entry the way the migration would write it, and the audit names it (`inbox.unmigrated`), as it names an unknown category and every `unsorted` entry.
|
|
187
225
|
- Lessons are entries of the category `lesson`, their kind the detail. `gcf board --json`: `inbox[]` is `{id, category, detail, origin, later, item, at, title, text, attachments?}`; `inbox_repeats` lists each lesson kind met in 3 items or more, Later left out; each question carries `category` (`decision` for D1, `assumption` for D2); `domain.sections.inbox` carries `groups` and `categories`; an item's `lessons` counts only the category `lesson`.
|
|
188
226
|
- **Ids from titles** (Board-PRD decision 34). `item new [id] --title <text>`: without an id the core makes it from the title (Cyrillic transliterated, about four words and 32 characters, `-2`, `-3`… when the id was ever used, `<type>-<n>` when nothing is left) and prints `<id> gcf/work/<id>.md` first; `--from` gives the title too. `item new` always holds the main lock. `--dry-run` prints the id and writes nothing. `--parent` takes one live feature. `design save --kind unit --system <s> --type <t> --title <text>` with no name makes a new unit's id the same way and prints it first (domain move `design-new-unit`). A pillar has an optional `name`; a new pillar's id comes from it; a changed name renames the id after the save, in a second commit `gcf(pillar): rename <a> → <b>`.
|
|
@@ -280,7 +318,7 @@ Update together with gcf-harness-aif-ext 0.12.0 and gcf-board 0.6.0 (docs/Sessio
|
|
|
280
318
|
- `gcf verify` kills a `dod.test` or `dod.build` command that runs longer than `dod_timeout_min` minutes (60 by default, 0 for no limit), with every process it started, and fails the DoD: `dod.test timed out after 60 min (dod_timeout_min), killed: <command>`. Before, a hung engine held the check, and a run with it, for hours.
|
|
281
319
|
- `gcf prioritize` takes an item made with `gcf item new` and never committed, and commits it with its rank. Before, it refused the item as an uncommitted file. An item with uncommitted edits is still refused.
|
|
282
320
|
- `gcf playtest session` lists the items awaiting a decision in a section of their own: before, they were missing from the sheet.
|
|
283
|
-
- `plan_passes` defaults to 3 (was 5). A plan converges when
|
|
321
|
+
- `plan_passes` defaults to 3 (was 5). A plan converges when its last review pass finds nothing blocking (Methodology §5.2.1); a project that sets `plan_passes` keeps its value.
|
|
284
322
|
|
|
285
323
|
### 0.7.9
|
|
286
324
|
|
|
@@ -331,7 +369,7 @@ Update together with gcf-harness-aif-ext 0.12.0 and gcf-board 0.6.0 (docs/Sessio
|
|
|
331
369
|
|
|
332
370
|
- `gcf init` restamps an unplayed item frozen by 0.4.0 (the hash of its body alone), so its `live:` is guarded too — only while that `live:` is still the one its close committed. Commit the restamped items it names.
|
|
333
371
|
- While a harness run drives an item (`.gcf-session/runs/<id>.lock` held by a live process), the core refuses any move on it, `/gcf-run` on main included, and `gcf integrate` leaves its branch unmerged; the audit and the board read it as `running`. A lock left by a run that died is taken over. `gcf board --json` reports WIP as `{count, limit, over}`.
|
|
334
|
-
- The audit reads the `until:` of every D2 record in `open-questions.md`: a fired one is a stale until record, and the assumption goes back to the human.
|
|
372
|
+
- The audit reads the `until:` of every D2 record in `open-questions.md`: a fired one is a stale until record, and the assumption goes back to the human (0.34.0 refines this, see there).
|
|
335
373
|
- `gcf design check` reads the text of an HTML comment in a unit or an item for forbidden words (M4) and numbers (M5); a `gcf:allow` pragma and a tool's marked region are skipped.
|
|
336
374
|
- An unplayed tune or kill keeps a systems-doc line the doc held before the item's close.
|
|
337
375
|
- `gcf playtest session` groups the checklist by subsystem and warns when the build is not of main: the remote's default branch when the clone knows it, else any branch but a detached HEAD or a `work/` one.
|
|
@@ -418,8 +456,9 @@ Removal leaves `gcf/` and `cnlp.config.json` in place. It also leaves the pre-co
|
|
|
418
456
|
|
|
419
457
|
**A deterministic core** behind `ai-factory gcf …`:
|
|
420
458
|
|
|
421
|
-
- work items — `take <ids...> [--feature <id>]`, `item new [id] --title <text> --type <type>`, `submit`, `close`, `verdict <id> keep|tune|kill`, `decide <id> go|no|wontfix|known|close|continue|kill`, `block`, `unblock`, `drop`, `done`, `where`, `item amend <id> [--text <blocks>] [--add <lines>] [--title <text>] [--from <in-
|
|
459
|
+
- work items — `take <ids...> [--feature <id>]`, `item new [id] --title <text> --type <type>`, `submit`, `close`, `verdict <id> keep|tune|kill`, `decide <id> go|no|wontfix|known|close|continue|kill`, `block`, `unblock`, `drop`, `done`, `where`, `item amend <id> [--text <blocks>] [--add <lines>] [--title <text>] [--from <in-ids>] [--proposal <key>] [--dry-run --json]`, `inbox drop <in-ids...> [--reason done|duplicate|noise|reported|amended] [--of <id|hash>]`, `inbox later <in-ids...> [--off]`
|
|
422
460
|
- inbox and snapshot — `inbox [add <text> [--category <c>] [--detail <word>] | add --text-file <path|-> | category <in-id> | later <in-id> [--off] | categories | drop <in-id> | migrate]`, `live`, `live check`
|
|
461
|
+
- lessons and rules (Lessons-Review-Plan d5, d6; the human's moves, refused to a run and a board thread, proposed by a chat agent) — `lessons report --from <in-ids> --to framework|template (--dry-run [--json] | --commit)`: a file a lesson in `<GCF_BOARD_HOME or ~/.gcf>/lessons/<to>/<project>-<in-id>.md`, the entries erased (`reported`) in one commit; `rules [--json]` reads `AGENTS.md` and `CLAUDE.md` (`{files: [{file, text, base, bytes, block, imports_agents, dirty}], own_rules}`); `rules add --text <line> [--from <in-ids>] [--base <hash>] (--dry-run | --commit)` puts one line under `## Project rules`, outside GCF's block, of `AGENTS.md` and of a `CLAUDE.md` that does not import it (refused: a repeat, several lines, a marker, a leading `@` or `#`, a code fence); `rules save --file AGENTS.md|CLAUDE.md --base <hash> --text-file - (--dry-run | --commit)` saves the whole file in its own line endings, refused for a stale base or a changed GCF block. A file with edits not committed goes into the commit whole; `--dry-run` shows its diff from HEAD. `--proposal <thread>/<id>` on `rules add` and `lessons report`; `rules diff --from <hash> --to <hash> [--names] [--json]` (a read, hex hashes only; `--names`: paths alone, no patch) gives what the commits `from..to` changed in the agent's instructions — `AGENTS.md`, `CLAUDE.md`, `.claude/`, `.ai-factory/` without `plans/` and `archive/` (`INSTRUCTION_PATHS`) — no merges, first parent only, so a worktree's `gcf(sync)` of main is not its own, nor a move of the operator's (`GCF-By: operator`) (`{from, to, paths, uncommitted, commits: [{sha, subject, diff}]}`; the harness names them in `run_end.instructions`)
|
|
423
462
|
- runs and play — `context`, `plan check <id>`, `verify <id>`, `playtest`, `playtest session`, `playtest status`, `playtest devices`, `play-notes`, `item show`, `audit`
|
|
424
463
|
- project phase and board — `phase`, `phase next [--reason]`, `board [--json]` (everything a board shows, in one read)
|
|
425
464
|
- branches — `worktree <id>`, `verify <id> --mark`, `integrate [<id>...]`: a run in its own worktree, the `GCF-Verified` trailer, the merge into main and the move; every committed move holds `.gcf-session/main.lock`
|
|
@@ -436,7 +475,7 @@ Every move of an item makes one commit `gcf(<move>): <id>` with every file it to
|
|
|
436
475
|
**CNLP profiles and a work item schema** for the documents the agent reads — one profile per unit type, shared by a unit and its item — plus injections into `aif-plan`, `aif-review` and `aif-commit`.
|
|
437
476
|
|
|
438
477
|
|
|
439
|
-
Review passes (Hold-Plan, Operator-Tasks-Plan t2): `review.plan` and `review.code` of `gcf/config.yml` are lists of rows `{agent: claude|codex, model, effort}` (`gcf config set review.plan --value "claude:opus:high, codex:astra:medium"`, `none`: no review; defaults Codex astra·
|
|
478
|
+
Review passes (Hold-Plan, Operator-Tasks-Plan t2): `review.plan` and `review.code` of `gcf/config.yml` are lists of rows `{agent: claude|codex, model, effort}` (`gcf config set review.plan --value "claude:opus:high, codex:astra:medium"`, `none`: no review; defaults Codex astra·high and fable·medium ×2 for the plan (the first pass is a full review, the next ones check the author's fixes: `gcf plan pass --findings`), Codex sol·high and opus·high for the code); row k makes pass k, the next only while the last found something blocking, of any verdict (accepted, rejected or downgraded). A row that cannot make its pass gives way to a model of its size (Fallback-Fidelity-Plan d2): Codex medium to fable·medium then opus·high, Codex high and fable·medium to opus·high, any other to none; `gcf board --json` carries the rules as `review_fallback`. `gcf review row <id> --kind plan|code --pass k [--diff] --json` names the row and its `fallback` (and writes the code's diff for the reviewer to `.gcf-session/review/<id>.diff`: from where the card's work started to the disk, new files in, `gcf/` and `no_diff` out). `gcf plan pass <id> --blocking N --polish M [--by agent:model:effort [--for agent:model:effort]] [--kind code]` records each pass (`, for` the row, when someone else made it: `self` or a stand-in) in the plan's `## Passes` (the code's `## Reviews`) and answers continue, converged or stalled; `gcf review stall <id> --kind plan|code` blocks a review that does not converge (`stall.plan|code`; a card's old `stall` is the plan's), and the human answers with `gcf review more <id> --passes N` or `gcf review accept <id>`. `plan_passes` and `plan_review:` are gone: `gcf audit` names the one `config set review.plan` command that moves them, and that set drops them.
|
|
440
479
|
|
|
441
480
|
## Docs
|
|
442
481
|
|