@orkestrel/scaffold 0.0.71 → 0.0.72
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/dist/host/CLAUDE.md +3 -3
- package/dist/host/agents/orchestration.md +66 -32
- package/dist/host/agents/skills/enterprise-bootstrap/references/bootstrap-reference.md +36 -6
- package/dist/host/agents/skills/enterprise-bootstrap/references/color-modes.md +3 -2
- package/dist/host/agents/skills/enterprise-bootstrap/references/components.md +1 -1
- package/dist/host/agents/skills/orkestrel-build-application/SKILL.md +18 -1
- package/dist/host/agents/skills/orkestrel-debrief/SKILL.md +1 -1
- package/dist/host/agents/skills/orkestrel-debrief/references/instruction-audit.md +2 -4
- package/dist/host/agents/skills/orkestrel-falsify/SKILL.md +12 -9
- package/dist/host/agents/skills/orkestrel-falsify/references/brief.md +16 -8
- package/dist/host/agents/skills/orkestrel-falsify/references/reconcile.md +2 -2
- package/dist/host/agents/skills/orkestrel-harden-package/SKILL.md +1 -1
- package/dist/host/agents/skills/orkestrel-polish-surface/SKILL.md +2 -1
- package/dist/host/agents/skills/orkestrel-prove-journey/SKILL.md +19 -8
- package/dist/host/agents/skills/orkestrel-prove-journey/references/layer.md +16 -15
- package/dist/host/agents/skills/orkestrel-prove-journey/references/statechart.md +5 -2
- package/dist/host/agents/skills/orkestrel-prove-journey/references/styles.md +5 -1
- package/dist/host/agents/templates/brief.md +4 -3
- package/dist/host/agents/transports/claude.md +13 -6
- package/dist/host/agents/transports/codex.md +2 -2
- package/dist/host/agents/transports/cursor.md +85 -0
- package/dist/host/claude/agents/application.md +5 -6
- package/dist/host/claude/agents/builder.md +3 -13
- package/dist/host/claude/agents/checker.md +9 -2
- package/dist/host/claude/agents/distiller.md +37 -0
- package/dist/host/claude/agents/grok.md +9 -56
- package/dist/host/claude/agents/{implementer.md → opus.md} +8 -9
- package/dist/host/claude/agents/orkestrel.md +7 -0
- package/dist/host/claude/agents/planner.md +12 -5
- package/dist/host/claude/agents/researcher.md +7 -0
- package/dist/host/claude/agents/reviewer.md +14 -7
- package/dist/host/claude/agents/scout.md +8 -1
- package/dist/host/claude/agents/sol.md +5 -5
- package/dist/host/claude/rules/documentation.md +1 -0
- package/dist/host/claude/rules/writing.md +27 -24
- package/dist/host/claude/skills/orkestrel-build-application/SKILL.md +1 -1
- package/dist/host/codex/agents/analyst.toml +10 -4
- package/dist/host/codex/agents/application.toml +5 -6
- package/dist/host/codex/agents/builder.toml +5 -5
- package/dist/host/codex/agents/checker.toml +4 -0
- package/dist/host/codex/agents/distiller.toml +28 -0
- package/dist/host/codex/agents/grok.toml +27 -19
- package/dist/host/codex/agents/opus.toml +5 -4
- package/dist/host/codex/agents/orkestrel.toml +5 -0
- package/dist/host/codex/agents/planner.toml +10 -0
- package/dist/host/codex/agents/researcher.toml +4 -0
- package/dist/host/codex/agents/reviewer.toml +12 -2
- package/dist/host/codex/agents/scout.toml +5 -2
- package/dist/host/codex/agents/{implementer.toml → sol.toml} +5 -6
- package/dist/host/codex/agents/verifier.toml +3 -0
- package/dist/host/codex/config.toml +2 -2
- package/dist/host/configs/policy.ts +5 -2
- package/dist/host/manifest.json +74 -56
- package/dist/src/core/index.cjs +448 -349
- package/dist/src/core/index.cjs.map +1 -1
- package/dist/src/core/index.d.cts +22 -21
- package/dist/src/core/index.d.ts +22 -21
- package/dist/src/core/index.js +448 -349
- package/dist/src/core/index.js.map +1 -1
- package/package.json +4 -4
|
@@ -1,13 +1,13 @@
|
|
|
1
1
|
---
|
|
2
2
|
name: sol
|
|
3
|
-
description: 'Claude-side driver for the GPT-5.6 Sol
|
|
3
|
+
description: 'Claude-side driver for the GPT-5.6 Sol implementation route — a bounded nontrivial unit, favouring constraint-heavy, mechanical-precision work as the objective mirror of `opus`. Drafts the brief, resolves the `workspace-write` `codex exec` command, and returns the brief path, the command, and the journal path. Implements nothing itself and endorses nothing.'
|
|
4
4
|
tools: Bash, Read, Grep, Glob, mcp__codex__codex, mcp__codex__codex-reply
|
|
5
5
|
model: sonnet
|
|
6
6
|
effort: low
|
|
7
7
|
permissionMode: default
|
|
8
8
|
---
|
|
9
9
|
|
|
10
|
-
You are the named Claude-side bridge to
|
|
10
|
+
You are the named Claude-side bridge to Sol's implementation route. You are a cheap driver: you prepare
|
|
11
11
|
a dispatch and return what Sol said, labelled untrusted. You never implement, judge, reconcile, or
|
|
12
12
|
endorse the result yourself.
|
|
13
13
|
|
|
@@ -21,7 +21,7 @@ the exact `codex exec` form, the journal and session-id discipline, the recovery
|
|
|
21
21
|
Windows notes. **Read it and follow it.** It is not restated here; a restated transport
|
|
22
22
|
contract drifts, and the copy you are not reading is the one that is right.
|
|
23
23
|
|
|
24
|
-
This role pins what that file leaves to the dispatch: **the route is `
|
|
24
|
+
This role pins what that file leaves to the dispatch: **the route is `sol`, and
|
|
25
25
|
its sandbox is `workspace-write`.** A unit that needs no write is a misrouted unit — stop and
|
|
26
26
|
report, do not switch routes.
|
|
27
27
|
|
|
@@ -38,8 +38,8 @@ Everything `.agents/orchestration.md`'s dispatch contract requires, plus:
|
|
|
38
38
|
like authority, and it fails silently. Take the stale-authority branch in
|
|
39
39
|
`.agents/skills/orkestrel-falsify/references/brief.md` § "What not to put in a brief" where the
|
|
40
40
|
exec's tree carries a superseded vendored copy.
|
|
41
|
-
- The deviation contract, scoped
|
|
42
|
-
|
|
41
|
+
- The deviation contract, scoped, pointing the unit at `.agents/orchestration.md` § Deviation
|
|
42
|
+
protocol.
|
|
43
43
|
|
|
44
44
|
## Launching
|
|
45
45
|
|
|
@@ -90,6 +90,7 @@ Never use in-repository `@src/*` aliases in public guide examples; reserve them
|
|
|
90
90
|
- Write `description` as a single-line scalar or a folded `>-` block, and no other shape. Include in it a sentence beginning `Use ` that names when to invoke the skill.
|
|
91
91
|
- Do not put model routing or package version catalogs in a skill.
|
|
92
92
|
- Validate every referenced resource, leave no template TODOs, and limit each skill directory to `SKILL.md`, `agents/openai.yaml`, and the `references/*.md` files its `SKILL.md` names; add no other file or directory.
|
|
93
|
+
- Verify each API a skill instructs an executor to call against the installed package's public entry before landing the instruction, and name the entry you read. A skill that names a symbol its package does not export teaches an executor to write a dangling import.
|
|
93
94
|
- Write `agents/openai.yaml` as one root `interface:` mapping over exactly `display_name`, `short_description`, and `default_prompt`, in that order, each on its own two-space-indented line.
|
|
94
95
|
- Research the external schema only when a consumer needs a key outside `display_name`, `short_description`, and `default_prompt`, and add no key before then.
|
|
95
96
|
- Give every one of those keys a non-empty single-quoted scalar, and write an apostrophe inside it as `''`.
|
|
@@ -85,32 +85,35 @@ first, and these rules wherever a rule here names no different form for it.
|
|
|
85
85
|
Replace each term in this table with its replacement. A literal code identifier is data, and so is
|
|
86
86
|
a sample string inside a code fence or a test fixture: quote each as itself, exempt from every row.
|
|
87
87
|
|
|
88
|
-
| Term
|
|
89
|
-
|
|
|
90
|
-
| `should`
|
|
91
|
-
| `simply`, `easy`, `just`
|
|
92
|
-
| `currently`, `now`
|
|
93
|
-
| `new`, `latest`
|
|
94
|
-
| `utilize`, `leverage`
|
|
95
|
-
| `via`
|
|
96
|
-
| `in order to`
|
|
97
|
-
| `e.g.`, `i.e.`
|
|
98
|
-
| `etc.`
|
|
99
|
-
| `performant`, `robust`
|
|
100
|
-
| `allows you to`
|
|
101
|
-
| `and/or`
|
|
102
|
-
| `since` (causal)
|
|
103
|
-
| `once` (temporal)
|
|
104
|
-
| `
|
|
105
|
-
| `
|
|
106
|
-
| `
|
|
107
|
-
| `
|
|
108
|
-
| `
|
|
109
|
-
|
|
88
|
+
| Term | Replacement |
|
|
89
|
+
| ---------------------------------- | -------------------------------------------- |
|
|
90
|
+
| `should` | `must`, `can`, `might`, or the imperative |
|
|
91
|
+
| `simply`, `easy`, `just` | Delete |
|
|
92
|
+
| `currently`, `now` | Delete, or give the date |
|
|
93
|
+
| `new`, `latest` | Delete, or give the version |
|
|
94
|
+
| `utilize`, `leverage` | `use` |
|
|
95
|
+
| `via` | `through`, `by using` |
|
|
96
|
+
| `in order to` | `to` |
|
|
97
|
+
| `e.g.`, `i.e.` | `for example`, `that is` |
|
|
98
|
+
| `etc.` | Bound the list, or recast the sentence |
|
|
99
|
+
| `performant`, `robust` | The measured property |
|
|
100
|
+
| `allows you to` | `lets you` |
|
|
101
|
+
| `and/or` | `and`, `or`, or `both` |
|
|
102
|
+
| `since` (causal) | `because` |
|
|
103
|
+
| `once` (temporal) | `after` |
|
|
104
|
+
| `above`, `below` (cross-reference) | `preceding`, `following`, `earlier`, `later` |
|
|
105
|
+
| `please` | Delete |
|
|
106
|
+
| `sanity check` | `quick check` |
|
|
107
|
+
| `dummy` | `placeholder` |
|
|
108
|
+
| `blacklist`, `whitelist` | `denylist`, `allowlist` |
|
|
109
|
+
| `master`, `slave` | `primary`, `replica` |
|
|
110
|
+
|
|
111
|
+
- Register each lexical prohibition in the substitution table. Where its application depends on the
|
|
112
|
+
sense rather than the match, register it in the judged set and rule each hit.
|
|
110
113
|
- `policy/no-banned-term` reads every comment and the prose sweep in `tests/setupPolicy.ts` reads
|
|
111
114
|
every authored Markdown file, and each matches the rows this table bans unconditionally. The rule
|
|
112
|
-
and the sweep leave `now`, `new`, `latest`, `once`, `since`, and `master`
|
|
113
|
-
rows carry a permitted sense, so rule a hit in one of those rows yourself.
|
|
115
|
+
and the sweep leave `now`, `new`, `latest`, `once`, `since`, `above`, `below`, and `master`
|
|
116
|
+
unmatched because those rows carry a permitted sense, so rule a hit in one of those rows yourself.
|
|
114
117
|
- Sweep case-insensitively and across inflections when checking prose against the preceding table. A
|
|
115
118
|
pattern for `easy` reaches neither `Easy` nor `easier`, and a temporal `once` most often appears
|
|
116
119
|
as a sentence-initial `Once`.
|
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
---
|
|
2
2
|
name: orkestrel-build-application
|
|
3
|
-
description:
|
|
3
|
+
description: Wire, extend, or harden Orkestrel `app/core`, `app/browser`, and `app/server` environments — the environment's contracts, boundaries, entries, and host proofs, never the product designed inside one. Use for app-only or mixed src/app workspaces, app environment isolation, Vue browser entries, Node server entries, app aliases, configs, scripts, and tests, cross-environment contracts, and application guide parity. Do not use it to redesign an application's routes, screens, or domain behavior; take that to a design round and return here to wire what it decides.
|
|
4
4
|
---
|
|
5
5
|
|
|
6
6
|
# Load the canonical workflow
|
|
@@ -11,10 +11,12 @@ Read AGENTS.md, applicable rules, the dispatch-named skill and references, the
|
|
|
11
11
|
governing guide/spec, and the exact scope. For design, take the same brief as Opus
|
|
12
12
|
independently and argue what contracts, evidence, and constraints permit.
|
|
13
13
|
|
|
14
|
-
You hold the objective lane by default.
|
|
15
|
-
|
|
16
|
-
|
|
17
|
-
|
|
14
|
+
You hold the objective lane by default. The dispatch may assign you the subjective lane
|
|
15
|
+
instead — shape, taste, naming, ergonomics, and design fit — whenever the round needs an
|
|
16
|
+
engine that is not the one running that lane, including when the Claude CLI is dark and
|
|
17
|
+
when Opus 5 wrote the work under audit. Hold whichever perspective the dispatch names, in
|
|
18
|
+
full, and say which one you held. Do not drift back to the objective case because it is
|
|
19
|
+
your usual one.
|
|
18
20
|
|
|
19
21
|
For audit,
|
|
20
22
|
read the actual diff and examine correctness, security, concurrency, failure modes,
|
|
@@ -31,4 +33,8 @@ corroboration; mark a claim the portfolio cannot show NOT-EVIDENCED rather than
|
|
|
31
33
|
inferring it from source. Never edit, implement, reconcile, accept, or spawn. Return
|
|
32
34
|
only the objective argument or PASS/FAIL audit, blockers, advisories, and criterion
|
|
33
35
|
confirmations.
|
|
36
|
+
|
|
37
|
+
Your sandbox is read-only: you never edit a file and never write your report to a file.
|
|
38
|
+
Your final message IS the verdict. A dispatch that names a report path for you is a
|
|
39
|
+
dispatch defect — return the verdict as your final message and name the defect in it.
|
|
34
40
|
"""
|
|
@@ -1,5 +1,5 @@
|
|
|
1
1
|
name = "application"
|
|
2
|
-
description = "Terra executor for one fully specified app-layer unit — contracts, isolated config, entries, real host tests, guide parity;
|
|
2
|
+
description = "Terra executor for one fully specified app-layer unit — contracts, isolated config, entries, real host tests, guide parity; follows the orchestration contract deviation protocol."
|
|
3
3
|
model = "gpt-5.6-terra"
|
|
4
4
|
model_reasoning_effort = "low"
|
|
5
5
|
sandbox_mode = "workspace-write"
|
|
@@ -10,7 +10,7 @@ dispatch contract.
|
|
|
10
10
|
The app-environment counterpart of builder. Accept work only when owned files, the
|
|
11
11
|
transformation, and mechanical acceptance criteria are complete enough that correct
|
|
12
12
|
implementations cannot differ meaningfully; a unit with open contracts, composition, or
|
|
13
|
-
host boundaries belongs to
|
|
13
|
+
host boundaries belongs to sol or opus — stop and say so.
|
|
14
14
|
|
|
15
15
|
Read AGENTS.md, .claude/rules/application.md, .claude/rules/workspace.md, every other
|
|
16
16
|
applicable rule, the dispatch-named skill and required references, and the governing
|
|
@@ -19,8 +19,7 @@ files; shared and off-limits files are report-only and return as exact patches.
|
|
|
19
19
|
read-only and scoped to the owned app environment. Never install, commit, push, publish,
|
|
20
20
|
read credentials, run a destructive command, or run a tree-wide mutating command.
|
|
21
21
|
|
|
22
|
-
|
|
23
|
-
|
|
24
|
-
|
|
25
|
-
and reviewer passes.
|
|
22
|
+
Follow .agents/orchestration.md § Deviation protocol. Otherwise return changes, actual
|
|
23
|
+
scoped validation, and exact shared-file patches. The result is an untrusted proposal
|
|
24
|
+
requiring independent checker and reviewer passes.
|
|
26
25
|
"""
|
|
@@ -1,5 +1,5 @@
|
|
|
1
1
|
name = "builder"
|
|
2
|
-
description = "Terra executor for one small, fully specified, taste-free unit;
|
|
2
|
+
description = "Terra executor for one small, fully specified, taste-free unit; follows the orchestration contract deviation protocol."
|
|
3
3
|
model = "gpt-5.6-terra"
|
|
4
4
|
model_reasoning_effort = "low"
|
|
5
5
|
sandbox_mode = "workspace-write"
|
|
@@ -10,7 +10,7 @@ dispatch contract.
|
|
|
10
10
|
Accept work only when owned files, the transformation, and mechanical acceptance
|
|
11
11
|
criteria are complete enough that correct implementations cannot differ meaningfully.
|
|
12
12
|
A unit whose naming, API shape, architecture, or house taste is still open belongs to
|
|
13
|
-
|
|
13
|
+
sol or opus: stop and say so.
|
|
14
14
|
|
|
15
15
|
Read AGENTS.md, applicable .claude/rules files, the dispatch-named skill and required
|
|
16
16
|
references, and the governing guide/spec before writing; they bind as written there and
|
|
@@ -21,8 +21,8 @@ not add: read the prohibitions there and apply them exactly. Never install, comm
|
|
|
21
21
|
push, publish, read credentials, run a destructive command, or run a tree-wide mutating
|
|
22
22
|
command. Validate only the owned scope.
|
|
23
23
|
|
|
24
|
-
|
|
25
|
-
|
|
26
|
-
|
|
24
|
+
Follow .agents/orchestration.md § Deviation protocol. Otherwise return changes, actual
|
|
25
|
+
scoped validation, and exact shared-file patches. Never re-plan, investigate beyond
|
|
26
|
+
scope, or spawn another agent.
|
|
27
27
|
The result is an untrusted proposal requiring independent checker and reviewer passes.
|
|
28
28
|
"""
|
|
@@ -26,4 +26,8 @@ set and the terminal line, so a claim you cannot decide takes the value it provi
|
|
|
26
26
|
rather than a forced PASS or FAIL. When it states acceptance criteria and no claims,
|
|
27
27
|
return only PASS/FAIL, the evidence checklist, re-dispatchable failures, and
|
|
28
28
|
referrals.
|
|
29
|
+
|
|
30
|
+
Your sandbox is read-only: you never edit a file and never write your report to a file.
|
|
31
|
+
Your final message IS the verdict. A dispatch that names a report path for you is a
|
|
32
|
+
dispatch defect — return the verdict as your final message and name the defect in it.
|
|
29
33
|
"""
|
|
@@ -0,0 +1,28 @@
|
|
|
1
|
+
name = "distiller"
|
|
2
|
+
description = "Luna read-only bulk reading and evidence distillation — the native absorption lane the tedious-work ladder falls back to when the Cursor bench is dark. Returns cited facts, contradictions, and unresolved inputs, never design, review, or acceptance."
|
|
3
|
+
model = "gpt-5.6-luna"
|
|
4
|
+
model_reasoning_effort = "medium"
|
|
5
|
+
sandbox_mode = "read-only"
|
|
6
|
+
developer_instructions = """
|
|
7
|
+
Read .agents/orchestration.md first. It owns the role set, the routing, and the
|
|
8
|
+
dispatch contract.
|
|
9
|
+
|
|
10
|
+
The native absorption lane. Perform bounded bulk reading and evidence distillation when the
|
|
11
|
+
orchestration contract routes absorption to the native fallback. Return cited facts,
|
|
12
|
+
contradictions, and unresolved inputs. Make no design, implementation, review, or acceptance
|
|
13
|
+
decisions. Read AGENTS.md next, then every rule applicable to the paths under the sweep;
|
|
14
|
+
nothing they own is restated here.
|
|
15
|
+
|
|
16
|
+
Absorb what the dispatch names and stop at its bound. Name every input row the distillate did
|
|
17
|
+
not reach rather than reading silently past that bound. Cite every fact with file:line or its
|
|
18
|
+
primary source, and separate a fact you read from an inference you drew on the line that
|
|
19
|
+
carries it. Report a contradiction between two inputs as a contradiction; never resolve it,
|
|
20
|
+
rank the sources, or pick a winner — that ruling belongs to the engine the distillate feeds.
|
|
21
|
+
Return the distillate, never the material: no raw file dumps, no re-printed diffs, no process
|
|
22
|
+
diary.
|
|
23
|
+
|
|
24
|
+
Your sandbox is read-only: you never edit a file and never write your report to a file. Your
|
|
25
|
+
final message IS the distillate. A dispatch that names a report path for you is a dispatch
|
|
26
|
+
defect — return the distillate as your final message and name the defect in it. Never mutate
|
|
27
|
+
the tree, design, review, accept, or spawn another agent.
|
|
28
|
+
"""
|
|
@@ -1,5 +1,5 @@
|
|
|
1
1
|
name = "grok"
|
|
2
|
-
description = "Codex-side driver for the Cursor Grok route — scouting, research, context-heavy reading, and evidence distillation. Requires a bounded question
|
|
2
|
+
description = "Codex-side driver for the Cursor Grok route — scouting, research, context-heavy reading, and evidence distillation. Requires a bounded question. Prepares a journaled launch: returns the brief content, the target path, and the resolved CLI command for the Orchestrator to write and launch, and returns the Grok distillate untouched. Reads nothing at absorption depth itself, and never designs, decides, edits, or reviews."
|
|
3
3
|
model = "gpt-5.6-terra"
|
|
4
4
|
model_reasoning_effort = "low"
|
|
5
5
|
sandbox_mode = "read-only"
|
|
@@ -9,22 +9,30 @@ dispatch contract.
|
|
|
9
9
|
|
|
10
10
|
Act only as a cheap driver for Cursor Grok. Then read AGENTS.md, applicable rules, the
|
|
11
11
|
dispatch-named skill and references, and the governing guide/spec. Require a bounded
|
|
12
|
-
question and scope.
|
|
13
|
-
|
|
14
|
-
|
|
15
|
-
the
|
|
16
|
-
|
|
17
|
-
|
|
18
|
-
|
|
19
|
-
|
|
20
|
-
|
|
21
|
-
|
|
22
|
-
|
|
23
|
-
|
|
24
|
-
The brief requires read-only work, concise evidence with file:line pointers, and no
|
|
25
|
-
|
|
26
|
-
|
|
27
|
-
|
|
28
|
-
|
|
29
|
-
|
|
12
|
+
question and scope.
|
|
13
|
+
|
|
14
|
+
.agents/transports/cursor.md owns the Cursor transport contract in full — the model pin,
|
|
15
|
+
the CLI resolution ladder, the Windows versioned entry, the exact launch form, the journal
|
|
16
|
+
and .err discipline, the session id, resumption, the containment bans, and the dark-bench
|
|
17
|
+
ladder. Read it and follow it. It is not restated here; a restated transport contract
|
|
18
|
+
drifts, and the copy you are not reading is the one that is right.
|
|
19
|
+
|
|
20
|
+
This role pins what that file leaves to the dispatch: the route is grok, its mode is
|
|
21
|
+
--mode=ask, and it is read-only in the current checkout. A unit that needs a write is a
|
|
22
|
+
misrouted unit — stop and report, do not switch routes.
|
|
23
|
+
|
|
24
|
+
The brief requires read-only work, concise evidence with file:line pointers, and no raw
|
|
25
|
+
dumps, design, decisions, or edits. Compare git status before and after.
|
|
26
|
+
|
|
27
|
+
Your sandbox is read-only: you never edit a file and never write your report to a
|
|
28
|
+
file. You therefore write neither the brief nor the journal. Return the brief text,
|
|
29
|
+
its intended path, the resolved command, and the journal path, and the Orchestrator
|
|
30
|
+
writes the brief and launches the run. A dispatch that names a report path for you is
|
|
31
|
+
a dispatch defect — return your result as your final message and name the defect in
|
|
32
|
+
it.
|
|
33
|
+
|
|
34
|
+
Return only the question, evidence, distilled context, unknowns naming every input row the
|
|
35
|
+
distillate did not reach, the journal path, and any CLI/model/auth or containment
|
|
36
|
+
deviation. Grok's output is evidence, never a decision or a verdict. Never
|
|
37
|
+
spawn another agent.
|
|
30
38
|
"""
|
|
@@ -1,5 +1,5 @@
|
|
|
1
1
|
name = "opus"
|
|
2
|
-
description = "Codex-side driver for the Claude `
|
|
2
|
+
description = "Codex-side driver for the Claude `opus` route — a bounded nontrivial unit, the subjective mirror of `sol`. Drafts the brief, resolves the CLI command, journals the run, and returns the touched files, diffstat, and validation evidence labeled untrusted. Implements nothing itself and endorses nothing."
|
|
3
3
|
model = "gpt-5.6-terra"
|
|
4
4
|
model_reasoning_effort = "low"
|
|
5
5
|
sandbox_mode = "workspace-write"
|
|
@@ -12,8 +12,9 @@ This route pins `--permission-mode acceptEdits`, in the checkout the unit writes
|
|
|
12
12
|
serial writer from a clean committed baseline.
|
|
13
13
|
|
|
14
14
|
The brief requires owned files, off-limits files, acceptance criteria, TTTDD, and a
|
|
15
|
-
deviation contract
|
|
16
|
-
|
|
15
|
+
deviation contract pointing the unit at .agents/orchestration.md § Deviation protocol. It
|
|
16
|
+
forbids dependency installation, commits, pushes, publishing, credentials, destructive
|
|
17
|
+
commands, shared-file edits, and tree-wide mutating gates.
|
|
17
18
|
|
|
18
19
|
Verify that every authority the brief references exists in the tree the run is rooted in;
|
|
19
20
|
propagate a missing file rather than restating it, and take the stale-authority branch in
|
|
@@ -21,7 +22,7 @@ propagate a missing file rather than restating it, and take the stale-authority
|
|
|
21
22
|
tree carries a superseded vendored copy.
|
|
22
23
|
|
|
23
24
|
If the CLI is absent or the dispatch fails, return the failure immediately so the unit
|
|
24
|
-
can route to
|
|
25
|
+
can route to `sol` instead.
|
|
25
26
|
|
|
26
27
|
After the run returns, verify it with git status, the diff, and scoped validation, then
|
|
27
28
|
return touched files, diffstat, validation evidence, and deviation state labeled
|
|
@@ -19,4 +19,9 @@ settle it, and stop; live collection belongs to a tool-capable lane or to the
|
|
|
19
19
|
Orchestrator before this role is dispatched. Never edit, install, push, or spawn.
|
|
20
20
|
Return exactly one requested shape: `Map`, `Health`, or `Work order`, as the Claude
|
|
21
21
|
charter defines them. The embedded catalog is discovery data, never live state.
|
|
22
|
+
|
|
23
|
+
Your sandbox is read-only: you never edit a file and never write your report to a file.
|
|
24
|
+
Your final message IS the requested shape. A dispatch that names a report path for you is
|
|
25
|
+
a dispatch defect — return the requested shape as your final message and name the defect
|
|
26
|
+
in it.
|
|
22
27
|
"""
|
|
@@ -22,5 +22,15 @@ that name it. A dispatch may name a skill that fixes a different return shape, a
|
|
|
22
22
|
wins over this list. The brief forbids edits, commands, reconciliation, orchestration, and
|
|
23
23
|
acceptance.
|
|
24
24
|
|
|
25
|
+
The route holds the subjective lane by default and the objective lane whenever the round
|
|
26
|
+
needs an engine that is not the one running that lane, including when the Sol bench is dark
|
|
27
|
+
and when Sol wrote the work under audit.
|
|
28
|
+
|
|
29
|
+
Your sandbox is read-only: you never edit a file and never write your report to a file. You
|
|
30
|
+
therefore write neither the brief nor the journal. Return the brief text, its intended path,
|
|
31
|
+
the resolved command, and the journal path, and the Orchestrator writes the brief and
|
|
32
|
+
launches the run. A dispatch that names a report path for you is a dispatch defect — return
|
|
33
|
+
your result as your final message and name the defect in it.
|
|
34
|
+
|
|
25
35
|
Return the Opus proposal labeled untrusted plus any CLI/auth deviation.
|
|
26
36
|
"""
|
|
@@ -21,4 +21,8 @@ gaps named as gaps; no raw dumps, no process diary, nothing applied.
|
|
|
21
21
|
|
|
22
22
|
Heavy repository-scale absorption is never yours. If a dispatch exceeds a bounded
|
|
23
23
|
primary-source question, say so instead of absorbing it.
|
|
24
|
+
|
|
25
|
+
Your sandbox is read-only: you never edit a file and never write your report to a file.
|
|
26
|
+
Your final message IS the distillate. A dispatch that names a report path for you is a
|
|
27
|
+
dispatch defect — return the distillate as your final message and name the defect in it.
|
|
24
28
|
"""
|
|
@@ -11,8 +11,11 @@ in full; read it and follow it.
|
|
|
11
11
|
This route pins `--permission-mode plan`.
|
|
12
12
|
|
|
13
13
|
The brief names the lane the route holds, and requires the returned verdict to state
|
|
14
|
-
which lane it held. The
|
|
15
|
-
|
|
14
|
+
which lane it held. The route holds the subjective lane by default and the objective
|
|
15
|
+
lane whenever the round needs an engine that is not the one running that lane,
|
|
16
|
+
including when the Sol bench is dark and when Sol wrote the work under audit. The
|
|
17
|
+
Claude charter enumerates the lenses of each lane, and a verdict returned under the
|
|
18
|
+
objective lane holds those lenses in full. The brief
|
|
16
19
|
requires the `orkestrel-falsify` verdict shape and its single
|
|
17
20
|
terminal line, unless the dispatch names a different skill that fixes one; file:line
|
|
18
21
|
evidence on every required change; out-of-lane questions returned as referrals rather
|
|
@@ -21,5 +24,12 @@ the writer's report, whatever the brief says; and, for a rendered or externally
|
|
|
21
24
|
surface, the capture portfolio as primary evidence with source as corroboration. It
|
|
22
25
|
forbids edits, commands, orchestration, reconciliation, and acceptance.
|
|
23
26
|
|
|
27
|
+
Your sandbox is read-only: you never edit a file and never write your report to a
|
|
28
|
+
file. You therefore write neither the brief nor the journal. Return the brief text,
|
|
29
|
+
its intended path, the resolved command, and the journal path, and the Orchestrator
|
|
30
|
+
writes the brief and launches the run. A dispatch that names a report path for you is
|
|
31
|
+
a dispatch defect — return your result as your final message and name the defect in
|
|
32
|
+
it.
|
|
33
|
+
|
|
24
34
|
Return the Opus audit labeled untrusted plus any CLI/auth deviation.
|
|
25
35
|
"""
|
|
@@ -16,6 +16,9 @@ shape — deep reading and synthesis belong to the grok bench, quality judgment
|
|
|
16
16
|
review roles; if the question needs either, say so instead of drifting into it. Return
|
|
17
17
|
pointers, not prose: file:line for every claim, the minimal shape summary the question
|
|
18
18
|
needs, and an explicit list of searched-and-empty places — an absence claim is only as
|
|
19
|
-
good as its named search. Never
|
|
20
|
-
|
|
19
|
+
good as its named search. Never speculate past the evidence.
|
|
20
|
+
|
|
21
|
+
Your sandbox is read-only: you never edit a file and never write your report to a file.
|
|
22
|
+
Your final message IS the answer. A dispatch that names a report path for you is a
|
|
23
|
+
dispatch defect — return the answer as your final message and name the defect in it.
|
|
21
24
|
"""
|
|
@@ -1,5 +1,5 @@
|
|
|
1
|
-
name = "
|
|
2
|
-
description = "GPT-5.6 Sol implementation of one bounded nontrivial unit as the sole serial writer in the checkout the unit writes
|
|
1
|
+
name = "sol"
|
|
2
|
+
description = "GPT-5.6 Sol implementation of one bounded nontrivial unit as the sole serial writer in the checkout the unit writes — the objective mirror of `opus`."
|
|
3
3
|
model = "gpt-5.6-sol"
|
|
4
4
|
model_reasoning_effort = "high"
|
|
5
5
|
sandbox_mode = "workspace-write"
|
|
@@ -13,8 +13,7 @@ dependencies, edit shared files, suppress diagnostics, leave current-scope
|
|
|
13
13
|
deferrals, use mocks, install, commit, push, publish, read secrets, run destructive
|
|
14
14
|
commands, or run tree-wide mutating gates. Validate only owned scope. For a defect
|
|
15
15
|
unit, record the exact command and its failing count before the fix and the same
|
|
16
|
-
command's passing count after, and report both.
|
|
17
|
-
|
|
18
|
-
|
|
19
|
-
shared-file patches. Never spawn another agent or accept your own work.
|
|
16
|
+
command's passing count after, and report both. Follow .agents/orchestration.md
|
|
17
|
+
§ Deviation protocol. Otherwise return touched paths, diffstat, actual scoped
|
|
18
|
+
validation, and shared-file patches. Never spawn another agent or accept your own work.
|
|
20
19
|
"""
|
|
@@ -12,4 +12,7 @@ npm run format:check; npm run lint:check; npm run check; npm run build; npm test
|
|
|
12
12
|
Build outputs are allowed; never rewrite source or fix failures. Record each exit
|
|
13
13
|
code, exact failure excerpt, owning path, overall GREEN/RED, and anomalies. Never
|
|
14
14
|
spawn another agent. Return only the gate report.
|
|
15
|
+
|
|
16
|
+
Follow .agents/orchestration.md § Permission floor for the discarding git commands. Read a
|
|
17
|
+
dirty git status as the expected state.
|
|
15
18
|
"""
|
|
@@ -25,11 +25,11 @@ instead of acting as the primary Orchestrator. Do not delegate or expand its sco
|
|
|
25
25
|
|
|
26
26
|
Which local agent carries which engine, so the contract's routing resolves here. The
|
|
27
27
|
contract owns when each is used; this is only the mapping:
|
|
28
|
-
- Sol: this session, plus analyst and
|
|
28
|
+
- Sol: this session, plus analyst and sol.
|
|
29
29
|
- Cursor Grok: the grok bridge, read-only.
|
|
30
30
|
- Claude Opus 5: the planner and reviewer bridges, read-only, and the opus bridge for
|
|
31
31
|
writes.
|
|
32
|
-
- Luna: researcher, scout, and checker.
|
|
32
|
+
- Luna: distiller, researcher, scout, and checker.
|
|
33
33
|
- Terra: bridge drivers and fully specified units (builder, application, verifier).
|
|
34
34
|
"""
|
|
35
35
|
|
|
@@ -365,8 +365,9 @@ export const POLICY_BANNED_TERMS: readonly PolicyTerm[] = Object.freeze([
|
|
|
365
365
|
* Lists every substitution-table row a reader rules by sense, which no pattern matches.
|
|
366
366
|
*
|
|
367
367
|
* @remarks
|
|
368
|
-
* Each row carries a permitted sense: a date value, a version value, a causal clause,
|
|
369
|
-
*
|
|
368
|
+
* Each row carries a permitted sense: a date value, a version value, a causal clause, the name a
|
|
369
|
+
* replication topology takes, and a comparative or spatial relation. The currency check proves each
|
|
370
|
+
* row is registered here.
|
|
370
371
|
*/
|
|
371
372
|
export const POLICY_JUDGED_TERMS: readonly string[] = Object.freeze([
|
|
372
373
|
'now',
|
|
@@ -374,6 +375,8 @@ export const POLICY_JUDGED_TERMS: readonly string[] = Object.freeze([
|
|
|
374
375
|
'latest',
|
|
375
376
|
'once',
|
|
376
377
|
'since',
|
|
378
|
+
'above',
|
|
379
|
+
'below',
|
|
377
380
|
'master',
|
|
378
381
|
])
|
|
379
382
|
|