pantheon-opencode 1.6.3-beta.15 → 1.6.3-beta.17
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/CHANGELOG.md +14 -0
- package/README.md +1 -1
- package/README.pt-BR.md +1 -1
- package/package.json +1 -1
- package/plugin.json +2 -2
- package/pyproject.toml +1 -1
- package/scripts/install/interactive.mjs +1 -1
- package/scripts/jev-live-probe.mjs +2 -2
- package/skills-lock.json +1 -0
- package/src/agents/athena.md +2 -0
- package/src/agents/nyx.md +12 -15
- package/src/agents/zeus.md +21 -0
- package/src/pantheon/jev-decision-tool.ts +6 -2
- package/src/pantheon/jev-decisions.ts +8 -4
- package/src/pantheon/jev-risk-gate.ts +69 -6
- package/src/pantheon/pantheon-decision-catalog.ts +6 -2
- package/src/plugin-v2.ts +25 -5
- package/src/plugins/tui/dist/tui-v2.js +3 -3
- package/src/plugins/tui/dist/tui.js +1 -1
- package/src/plugins/tui/package-lock.json +2 -2
- package/src/plugins/tui/package.json +1 -1
- package/src/routing.yml +1 -0
- package/src/skills/README.md +1 -1
- package/src/skills/prompt-refinery/SKILL.md +41 -0
package/CHANGELOG.md
CHANGED
|
@@ -10,6 +10,20 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/).
|
|
|
10
10
|
<!-- Add new changes here. Running `node scripts/versioning.mjs apply` will
|
|
11
11
|
move this section to a versioned entry and reset the template below. -->
|
|
12
12
|
|
|
13
|
+
## [v1.6.3-beta.17] - 2026-11-05
|
|
14
|
+
|
|
15
|
+
## 🆕 What's New
|
|
16
|
+
- The J1 gate is now a ladder: off / shadow / warn / enforce (PANTHEON_J1_RISK_SCORE). Enforce awaits the bounded gate and DENIES a high-risk tool call in-session, with the threshold configurable (PANTHEON_J1_ENFORCE_THRESHOLD, default 0.8). Shadow stays the default-on behavior.
|
|
17
|
+
- CALLER_NOT_ALLOWED refusals now teach the fix: allowlist the caller via PANTHEON_JEV_SPECIALIST_CALLERS or ask Zeus to route.
|
|
18
|
+
- Plan persistence contract: Athena returns the plan as her response; Zeus persists PLAN.md — no more refused write attempts on deepwork plans.
|
|
19
|
+
|
|
20
|
+
## [v1.6.3-beta.16] - 2026-11-05
|
|
21
|
+
|
|
22
|
+
## 🆕 What's New
|
|
23
|
+
- `prompt-refinery` skill: a vague multi-part request is refined into a structured contract (objective, slices, acceptance criteria, open questions with defaults) shown for confirm-or-adjust BEFORE executing — one planning turn on the session model.
|
|
24
|
+
- Skills mode offers the full local catalog + host skills — the lexical prefilter could drop the exact skill you described (codemap → none 0.88). candidateSource still reports the pool.
|
|
25
|
+
- nyx returns to observability & cost analytics (owner intent), prometheus drops those words — the routing you expected now matches the catalog.
|
|
26
|
+
|
|
13
27
|
## [v1.6.3-beta.15] - 2026-11-05
|
|
14
28
|
|
|
15
29
|
## 🆕 What's New
|
package/README.md
CHANGED
|
@@ -93,7 +93,7 @@ see [Quick Start](docs/QUICKSTART.md).
|
|
|
93
93
|
|
|
94
94
|
- **14 specialized agents** with one orchestrator, implementation specialists,
|
|
95
95
|
a quality gate, and supporting roles.
|
|
96
|
-
- **
|
|
96
|
+
- **21 skills** and reusable instructions for planning, building, reviewing,
|
|
97
97
|
and documenting work.
|
|
98
98
|
- **5 local MCP servers** for project memory, persistence, resources, confined
|
|
99
99
|
script execution, and vision.
|
package/README.pt-BR.md
CHANGED
|
@@ -71,7 +71,7 @@ apenas — `init` → reiniciar → `/dispatch` → primeiro resultado — veja
|
|
|
71
71
|
|
|
72
72
|
- **14 agentes especializados**, com um orquestrador, especialistas de
|
|
73
73
|
implementação, um gate de qualidade e papéis de apoio.
|
|
74
|
-
- **
|
|
74
|
+
- **21 skills** e instruções reutilizáveis para planejar, construir, revisar e
|
|
75
75
|
documentar.
|
|
76
76
|
- **5 servidores MCP locais** para memória de projeto, persistência, recursos,
|
|
77
77
|
execução confinada de scripts e visão.
|
package/package.json
CHANGED
package/plugin.json
CHANGED
|
@@ -1,9 +1,9 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "pantheon",
|
|
3
3
|
"publisher": "ils15",
|
|
4
|
-
"version": "1.6.3-beta.
|
|
4
|
+
"version": "1.6.3-beta.17",
|
|
5
5
|
"displayName": "Pantheon",
|
|
6
|
-
"description": "Conductor-Delegate orchestration platform with 14 specialized agents,
|
|
6
|
+
"description": "Conductor-Delegate orchestration platform with 14 specialized agents, 21 skills, and OpenCode support.",
|
|
7
7
|
"agents": "./src/agents",
|
|
8
8
|
"skills": "./src/skills",
|
|
9
9
|
"instructions": "./src/instructions",
|
package/pyproject.toml
CHANGED
|
@@ -4,7 +4,7 @@ build-backend = "setuptools.backends._legacy:_Backend"
|
|
|
4
4
|
|
|
5
5
|
[project]
|
|
6
6
|
name = "pantheon"
|
|
7
|
-
version = "1.6.3-beta.
|
|
7
|
+
version = "1.6.3-beta.17"
|
|
8
8
|
description = "Multi-agent orchestration framework"
|
|
9
9
|
requires-python = ">=3.11"
|
|
10
10
|
# Python sources under src/mcp/ import fastmcp/pydantic at runtime, but those
|
|
@@ -93,7 +93,7 @@ export async function runInteractiveInstall({
|
|
|
93
93
|
|
|
94
94
|
const componentOptions = [
|
|
95
95
|
{ label: 'agents', value: 'agents', hint: 'Agent configurations and routing (14 agents)' },
|
|
96
|
-
{ label: 'skills', value: 'skills', hint: `Reusable skill workflows (${
|
|
96
|
+
{ label: 'skills', value: 'skills', hint: `Reusable skill workflows (${21} skills)` },
|
|
97
97
|
{ label: 'instructions', value: 'instructions', hint: 'AGENTS.md and instructional files' },
|
|
98
98
|
{ label: 'prompts', value: 'prompts', hint: 'Prompt templates for agents' },
|
|
99
99
|
{ label: 'commands', value: 'commands', hint: 'OpenCode command shortcuts' },
|
|
@@ -73,14 +73,14 @@ function callDecisions(modelRef, state) {
|
|
|
73
73
|
{ value: 'athena', description: 'planning and architecture' },
|
|
74
74
|
{ value: 'themis', description: 'security, review and compliance gates' },
|
|
75
75
|
{ value: 'hermes', description: 'backend implementation' },
|
|
76
|
-
{ value: 'prometheus', description: 'infrastructure, CI and
|
|
76
|
+
{ value: 'prometheus', description: 'infrastructure, Docker, CI and model providers' },
|
|
77
77
|
{ value: 'hephaestus', description: 'AI tooling and model integration' },
|
|
78
78
|
{ value: 'demeter', description: 'databases and data layer' },
|
|
79
79
|
{ value: 'gaia', description: 'remote sensing and geospatial' },
|
|
80
80
|
{ value: 'iris', description: 'GitHub operations and automation' },
|
|
81
81
|
{ value: 'aphrodite', description: 'UI and frontend craft' },
|
|
82
82
|
{ value: 'mnemosyne', description: 'memory curation and recall' },
|
|
83
|
-
{ value: 'nyx', description: '
|
|
83
|
+
{ value: 'nyx', description: 'observability, tracing and cost analytics; chaos testing' },
|
|
84
84
|
{ value: 'talos', description: 'hotfix express lane - small direct fixes, no ceremony' },
|
|
85
85
|
{ value: 'none', description: 'no specialist fits' },
|
|
86
86
|
],
|
package/skills-lock.json
CHANGED
|
@@ -10,6 +10,7 @@
|
|
|
10
10
|
"src/skills/incremental-implementation/SKILL.md": "7065ddb45da3c6b6751de07f290e92e6cb072e949a2e3aaae13ee7762f0a70fc",
|
|
11
11
|
"src/skills/loop-engineering/SKILL.md": "f804e545821b9cca8fe0ec517307ed902446f28c2ca8ff9194c0f987cf7229e3",
|
|
12
12
|
"src/skills/memory-bank/SKILL.md": "6c27623e5588b7ba44fbd0ea88e5061ed36894e1c7cbc88e50f003c2b463d9ff",
|
|
13
|
+
"src/skills/prompt-refinery/SKILL.md": "d2ed35b1d74ee1caef07b572ee7771e19019a9bc6ad40ea71c4b32dbccd256d9",
|
|
13
14
|
"src/skills/reflect/SKILL.md": "4c7608c2a55cb8a50ac90fc0e2446bdf25e7914b1748cb7731c311d13c29dfdd",
|
|
14
15
|
"src/skills/security-hardening/SKILL.md": "6d637e122421f7bc3b751f04854ad343a5a5d35249458bc8939233138fd3f20d",
|
|
15
16
|
"src/skills/session-goal/SKILL.md": "2911157c8e020f62ca015a73a46f5427f41da71c83e624d6a1cecf7d64c64aea",
|
package/src/agents/athena.md
CHANGED
|
@@ -248,3 +248,5 @@ For external docs/specs, use `webfetch` (see `internet-search` skill for pattern
|
|
|
248
248
|
|
|
249
249
|
## Skills
|
|
250
250
|
Planejamento: `spec-driven-development`, `artifact-management`
|
|
251
|
+
|
|
252
|
+
Your plan is your DELIVERABLE: return it fully in your response. You cannot write files or persist PLAN.md — say so once and hand the plan back; Zeus persists it.
|
package/src/agents/nyx.md
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
---
|
|
2
2
|
name: nyx
|
|
3
|
-
description: "
|
|
3
|
+
description: "Observability & cost analytics specialist — OpenTelemetry tracing, per-agent cost dashboards, anomaly detection; also chaos and adversarial testing (fuzzing, hostile inputs, boundaries). Analysis-only: never fixes, routes through Zeus."
|
|
4
4
|
mode: all
|
|
5
5
|
reasoning_effort: low
|
|
6
6
|
permission:
|
|
@@ -61,22 +61,19 @@ agent.
|
|
|
61
61
|
|
|
62
62
|
## Core Capabilities
|
|
63
63
|
|
|
64
|
-
### 1.
|
|
65
|
-
-
|
|
66
|
-
|
|
67
|
-
-
|
|
68
|
-
- Distinguish crashes, hangs, silent corruption, and unvalidated accept
|
|
64
|
+
### 1. Observability & Tracing
|
|
65
|
+
- OpenTelemetry distributed tracing across services and agent pipelines
|
|
66
|
+
- Per-agent cost/latency dashboards (token usage attribution, spans)
|
|
67
|
+
- Anomaly detection: latency and spend outliers, error-rate cliffs
|
|
69
68
|
|
|
70
|
-
### 2.
|
|
71
|
-
-
|
|
72
|
-
|
|
73
|
-
- Complement Themis (security gate): Nyx attacks the RUNNING behavior
|
|
74
|
-
Themis reviews the code — send code-level findings to Zeus for Themis
|
|
69
|
+
### 2. Adversarial & Chaos (secondary)
|
|
70
|
+
- Fuzzing with malformed payloads; hostile-input boundary exploration
|
|
71
|
+
- Chaos scenarios for the monitored paths; findings feed the dashboards
|
|
72
|
+
- Complement Themis (security gate): Nyx attacks the RUNNING behavior
|
|
75
73
|
|
|
76
|
-
### 3.
|
|
77
|
-
-
|
|
78
|
-
|
|
79
|
-
- Report which boundaries the code defends and which it ignores
|
|
74
|
+
### 3. Cost & Metric Review
|
|
75
|
+
- Token/cost attribution per agent and per phase (feeds `pantheon_cost`)
|
|
76
|
+
- Alert thresholds and anomaly reports on the agent pipeline
|
|
80
77
|
|
|
81
78
|
## Handoffs
|
|
82
79
|
- **Discovery:** Send discovery needs and any findings to Zeus. Zeus may dispatch to targets; only root Athena/Hermes sessions may call @apollo directly. Child sessions must not dispatch another Pantheon agent.
|
package/src/agents/zeus.md
CHANGED
|
@@ -22,6 +22,7 @@ skills:
|
|
|
22
22
|
- agent-coordination
|
|
23
23
|
- artifact-management
|
|
24
24
|
- auto-continue
|
|
25
|
+
- prompt-refinery
|
|
25
26
|
- codemap
|
|
26
27
|
- context-compression
|
|
27
28
|
- incremental-implementation
|
|
@@ -93,6 +94,26 @@ call before firing it.
|
|
|
93
94
|
cheap and typed, an MCP call is the LAST resort for state the session
|
|
94
95
|
cannot reconstruct.
|
|
95
96
|
|
|
97
|
+
## Plan persistence (deepwork PLAN.md)
|
|
98
|
+
|
|
99
|
+
Athena is plan-only and cannot write files; a plan she produces returns as her
|
|
100
|
+
response. YOU persist it: write `.pantheon/deepwork/<slug>/PLAN.md` yourself
|
|
101
|
+
(or hand the write to the specialist that owns the artifact) right after her
|
|
102
|
+
plan lands. Never ask Athena to save, and never delegate persistence to her —
|
|
103
|
+
the field refusal (beta.16) was the frontmatter working as designed.
|
|
104
|
+
|
|
105
|
+
## Vague multi-part request? Refine into a contract first
|
|
106
|
+
|
|
107
|
+
When the user's request spans multiple concerns without explicit acceptance
|
|
108
|
+
criteria (refactor + fix + redesign in one paragraph), refine it into a
|
|
109
|
+
structured contract BEFORE executing: objective, numbered scope slices,
|
|
110
|
+
acceptance criteria per slice, deliverables, and open questions with your
|
|
111
|
+
recommended defaults. Load `skill: prompt-refinery` for the shape and rules.
|
|
112
|
+
Show the contract once for confirm-or-adjust — silence proceeds with the
|
|
113
|
+
stated defaults — then route the confirmed slices through the normal ladder.
|
|
114
|
+
One refinement turn on the session model; never a dialogue, never a Jev call
|
|
115
|
+
(decision models answer scalars, they do not write prose).
|
|
116
|
+
|
|
96
117
|
## Missing capability? Find a skill before inventing one
|
|
97
118
|
|
|
98
119
|
The skill-discovery ladder — installed pool first, ecosystem second, and the user always owns the install:
|
|
@@ -449,10 +449,14 @@ export async function runJevDecision(
|
|
|
449
449
|
? (toolContext as JevToolContext)
|
|
450
450
|
: undefined
|
|
451
451
|
const caller = context?.agent
|
|
452
|
-
if (typeof caller !== 'string' || caller.length === 0)
|
|
452
|
+
if (typeof caller !== 'string' || caller.length === 0)
|
|
453
|
+
return refusal('CALLER_NOT_ALLOWED', 'CALLER_NOT_ALLOWLISTED')
|
|
453
454
|
const allowlist = runtime.specialistCallers ?? []
|
|
454
455
|
if (caller !== 'zeus' && !allowlist.includes(caller)) {
|
|
455
|
-
return refusal(
|
|
456
|
+
return refusal(
|
|
457
|
+
'CALLER_NOT_ALLOWED',
|
|
458
|
+
'CALLER_NOT_ALLOWLISTED — allowlist this caller via PANTHEON_JEV_SPECIALIST_CALLERS (then restart) or ask Zeus to route',
|
|
459
|
+
)
|
|
456
460
|
}
|
|
457
461
|
|
|
458
462
|
const requested = (input ?? {}) as { mode?: unknown; questions?: unknown }
|
|
@@ -585,11 +585,15 @@ export function prepareJevPresetRequest(
|
|
|
585
585
|
return { ok: true, request: Object.freeze(request) }
|
|
586
586
|
}
|
|
587
587
|
|
|
588
|
-
|
|
588
|
+
// Field report beta.15 §skills: the lexical prefilter dropped the exact
|
|
589
|
+
// skill the state described (codemap → none 0.88 from a prefiltered pool).
|
|
590
|
+
// The local catalog is 12 entries and the choice bound is 24 — recall beats
|
|
591
|
+
// a filter that saves half of $0.00003. The full local catalog + host
|
|
592
|
+
// skills always ride; `candidateSource` keeps reporting which pool the
|
|
593
|
+
// caller got.
|
|
589
594
|
const hostHasSkills = Array.isArray(hostSkills) && hostSkills.length > 0
|
|
590
|
-
const candidateSource: JevSkillCandidateSource =
|
|
591
|
-
|
|
592
|
-
const pool = shortlist.length > 0 ? shortlist : CATALOG_SKILL_IDS
|
|
595
|
+
const candidateSource: JevSkillCandidateSource = hostHasSkills ? 'host-skills' : 'full-catalog'
|
|
596
|
+
const pool = CATALOG_SKILL_IDS
|
|
593
597
|
const descriptions = new Map(PANTHEON_SKILL_CATALOG.map((entry) => [entry.id, entry.description]))
|
|
594
598
|
const request = {
|
|
595
599
|
model: route.model,
|
|
@@ -120,12 +120,50 @@ const WRITE_TOOLS: ReadonlySet<string> = new Set([
|
|
|
120
120
|
'hashline_edit',
|
|
121
121
|
])
|
|
122
122
|
|
|
123
|
-
/**
|
|
123
|
+
/**
|
|
124
|
+
* The escalation ladder (field beta.14: the operator wants iterative
|
|
125
|
+
* escalation, not a binary flag). `on`/`1`/`true` map to `shadow` for
|
|
126
|
+
* back-compatibility with the shipped setup.
|
|
127
|
+
* off — nothing fires (hard zero)
|
|
128
|
+
* shadow — observe + record; invisible in the session (today's behavior)
|
|
129
|
+
* warn — RESERVED (needs a V2 client channel; currently behaves as shadow)
|
|
130
|
+
* enforce — an affirmative high-risk score DENIES the tool call (the host
|
|
131
|
+
* surfaces the block message in the session); gate unavailability
|
|
132
|
+
* never blocks (fail-open, documented)
|
|
133
|
+
*/
|
|
134
|
+
export type J1RiskGateMode = 'off' | 'shadow' | 'warn' | 'enforce'
|
|
135
|
+
|
|
136
|
+
export const J1_ENFORCE_THRESHOLD_ENV = 'PANTHEON_J1_ENFORCE_THRESHOLD'
|
|
137
|
+
export const J1_ENFORCE_THRESHOLD_DEFAULT = 0.8
|
|
138
|
+
|
|
139
|
+
export function j1ModeFromEnv(
|
|
140
|
+
env: Record<string, string | undefined> = process.env,
|
|
141
|
+
): J1RiskGateMode {
|
|
142
|
+
const raw = (env[J1_RISK_GATE_FLAG_ENV] ?? '').trim().toLowerCase()
|
|
143
|
+
if (raw === 'enforce') return 'enforce'
|
|
144
|
+
if (raw === 'warn') return 'warn'
|
|
145
|
+
return FLAG_ON.has(raw) ? 'shadow' : 'off'
|
|
146
|
+
}
|
|
147
|
+
|
|
148
|
+
/** Back-compatible: any non-off mode counts as enabled. */
|
|
124
149
|
export function j1RiskGateEnabledFromEnv(
|
|
125
150
|
env: Record<string, string | undefined> = process.env,
|
|
126
151
|
): boolean {
|
|
127
|
-
|
|
128
|
-
|
|
152
|
+
return j1ModeFromEnv(env) !== 'off'
|
|
153
|
+
}
|
|
154
|
+
|
|
155
|
+
/**
|
|
156
|
+
* The enforce-mode deny threshold (0..1). Invalid or out-of-range values fall
|
|
157
|
+
* back to the documented default — an operator typo must not disable the gate.
|
|
158
|
+
*/
|
|
159
|
+
export function j1EnforceThresholdFromEnv(
|
|
160
|
+
env: Record<string, string | undefined> = process.env,
|
|
161
|
+
): number {
|
|
162
|
+
const text = (env[J1_ENFORCE_THRESHOLD_ENV] ?? '').trim()
|
|
163
|
+
if (text === '') return J1_ENFORCE_THRESHOLD_DEFAULT
|
|
164
|
+
const raw = Number(text)
|
|
165
|
+
if (!Number.isFinite(raw) || raw < 0 || raw > 1) return J1_ENFORCE_THRESHOLD_DEFAULT
|
|
166
|
+
return raw
|
|
129
167
|
}
|
|
130
168
|
|
|
131
169
|
/**
|
|
@@ -408,12 +446,25 @@ function abortDeadline(signal: AbortSignal): Promise<'timeout'> {
|
|
|
408
446
|
* settles its own ledger reservation in the background (settlement of an
|
|
409
447
|
* already-fired call — not a retry, and it writes no record).
|
|
410
448
|
*/
|
|
449
|
+
export interface J1Verdict {
|
|
450
|
+
/** 'block' only in enforce mode with an affirmative high-risk score. */
|
|
451
|
+
block: boolean
|
|
452
|
+
riskScore: number | null
|
|
453
|
+
reason: string
|
|
454
|
+
}
|
|
455
|
+
|
|
411
456
|
export function createJ1RiskGateObserver(
|
|
412
|
-
options: J1RiskGateObserverOptions
|
|
413
|
-
|
|
457
|
+
options: J1RiskGateObserverOptions & {
|
|
458
|
+
/** enforce: the caller awaits and denies on { block: true }. */
|
|
459
|
+
mode?: J1RiskGateMode
|
|
460
|
+
threshold?: number
|
|
461
|
+
},
|
|
462
|
+
): (event: unknown) => Promise<void | J1Verdict> {
|
|
414
463
|
const timeoutMs = options.timeoutMs ?? J1_RISK_GATE_TIMEOUT_MS
|
|
415
464
|
const sink = options.sink
|
|
416
|
-
|
|
465
|
+
const enforce = options.mode === 'enforce'
|
|
466
|
+
const threshold = options.threshold ?? J1_ENFORCE_THRESHOLD_DEFAULT
|
|
467
|
+
return async (event: unknown): Promise<void | J1Verdict> => {
|
|
417
468
|
try {
|
|
418
469
|
const payload = (event ?? {}) as J1ExecuteBeforeEvent
|
|
419
470
|
if (!isRiskGateTarget(payload.tool)) return
|
|
@@ -467,6 +518,18 @@ export function createJ1RiskGateObserver(
|
|
|
467
518
|
}),
|
|
468
519
|
sink,
|
|
469
520
|
)
|
|
521
|
+
if (enforce) {
|
|
522
|
+
const risk = evaluation.riskScore
|
|
523
|
+
const block = risk !== null && risk >= threshold
|
|
524
|
+
return {
|
|
525
|
+
block,
|
|
526
|
+
riskScore: risk,
|
|
527
|
+
reason: block
|
|
528
|
+
? `J1 enforce: risco ${risk.toFixed(2)} >= threshold ${threshold}`
|
|
529
|
+
: 'allow',
|
|
530
|
+
}
|
|
531
|
+
}
|
|
532
|
+
return
|
|
470
533
|
} finally {
|
|
471
534
|
clearTimeout(timer)
|
|
472
535
|
}
|
|
@@ -53,10 +53,14 @@ export const PANTHEON_AGENT_CATALOG: readonly DecisionCatalogEntry[] = Object.fr
|
|
|
53
53
|
{ id: 'themis', description: 'Quality and security review gate.' },
|
|
54
54
|
{
|
|
55
55
|
id: 'prometheus',
|
|
56
|
-
description: 'Infrastructure, Docker, CI,
|
|
56
|
+
description: 'Infrastructure, Docker, CI, model providers and cloud costs.',
|
|
57
57
|
},
|
|
58
58
|
{ id: 'hephaestus', description: 'AI pipelines: LangChain, RAG, vector stores.' },
|
|
59
|
-
{
|
|
59
|
+
{
|
|
60
|
+
id: 'nyx',
|
|
61
|
+
description:
|
|
62
|
+
'Observability and cost analytics: tracing, per-agent metrics; chaos and fuzz testing.',
|
|
63
|
+
},
|
|
60
64
|
{ id: 'gaia', description: 'Remote sensing and geospatial analysis.' },
|
|
61
65
|
{ id: 'iris', description: 'GitHub branches, pull requests, issues, releases.' },
|
|
62
66
|
{ id: 'mnemosyne', description: 'Memory bank and ADR authoring on request.' },
|
package/src/plugin-v2.ts
CHANGED
|
@@ -126,7 +126,9 @@ import {
|
|
|
126
126
|
import {
|
|
127
127
|
createJ1RiskGateObserver,
|
|
128
128
|
isRiskGateTarget,
|
|
129
|
-
|
|
129
|
+
type J1Verdict,
|
|
130
|
+
j1EnforceThresholdFromEnv,
|
|
131
|
+
j1ModeFromEnv,
|
|
130
132
|
} from './pantheon/jev-risk-gate.ts'
|
|
131
133
|
import { DECISION_EGRESS_ACTION } from './pantheon/local-decision-egress-gate.ts'
|
|
132
134
|
import { applyDecisionRouterToPrompt } from './pantheon/local-decision-prompt.ts'
|
|
@@ -1984,14 +1986,17 @@ async function registerV2ToolHooks(
|
|
|
1984
1986
|
// ledger reservation, permission kill-switch) and records shadow records
|
|
1985
1987
|
// only — it is fired WITHOUT await inside the before-handler (fire-and-
|
|
1986
1988
|
// observe), self-bounds at 2s with an AbortController, and never retries.
|
|
1987
|
-
|
|
1988
|
-
|
|
1989
|
+
const j1Mode = j1ModeFromEnv()
|
|
1990
|
+
let j1Observer: ((event: unknown) => Promise<void | J1Verdict>) | undefined
|
|
1991
|
+
if (jevRuntime !== undefined && j1Mode !== 'off') {
|
|
1989
1992
|
const j1Sink = createTelemetrySink()
|
|
1990
1993
|
j1TelemetrySink = j1Sink
|
|
1991
1994
|
j1Observer = createJ1RiskGateObserver({
|
|
1992
1995
|
runtime: jevRuntime,
|
|
1996
|
+
mode: j1Mode,
|
|
1997
|
+
threshold: j1EnforceThresholdFromEnv(),
|
|
1993
1998
|
sink: (record) => {
|
|
1994
|
-
j1Sink.emit({ ...record, mode: 'shadow' })
|
|
1999
|
+
j1Sink.emit({ ...record, mode: j1Mode === 'enforce' ? 'enforce' : 'shadow' })
|
|
1995
2000
|
},
|
|
1996
2001
|
})
|
|
1997
2002
|
}
|
|
@@ -2148,7 +2153,22 @@ async function registerV2ToolHooks(
|
|
|
2148
2153
|
// the flag is off — the only cost of the flag-on check below is the
|
|
2149
2154
|
// truthiness read.
|
|
2150
2155
|
if (j1Observer !== undefined && isRiskGateTarget((event as { tool?: unknown }).tool)) {
|
|
2151
|
-
|
|
2156
|
+
// Ladder semantics: shadow/warn stay fire-and-observe (latency
|
|
2157
|
+
// unchanged); ENFORCE awaits (bounded 2s) and DENIES via throw —
|
|
2158
|
+
// the host surfaces the thrown message in the session, which is the
|
|
2159
|
+
// block's visible form. Unavailability never denies (fail-open by
|
|
2160
|
+
// design: a dead budget must not brick every shell call).
|
|
2161
|
+
if (j1Mode === 'enforce') {
|
|
2162
|
+
const verdict = (await j1Observer(event)) as J1Verdict | undefined
|
|
2163
|
+
if (verdict?.block === true) {
|
|
2164
|
+
throw new Error(
|
|
2165
|
+
`[J1 enforce] chamada bloqueada: risco ${verdict.riskScore?.toFixed(2)} ` +
|
|
2166
|
+
`>= threshold (ajuste PANTHEON_J1_ENFORCE_THRESHOLD ou rebaixe o modo para shadow/warn)`,
|
|
2167
|
+
)
|
|
2168
|
+
}
|
|
2169
|
+
} else {
|
|
2170
|
+
void j1Observer(event)
|
|
2171
|
+
}
|
|
2152
2172
|
}
|
|
2153
2173
|
// J3 turn-complexity shadow classification. Same fire-and-observe
|
|
2154
2174
|
// discipline as J1: the dispatch's latency is unchanged, the observer
|
|
@@ -367,7 +367,7 @@ async function readChangelogForBundle(bundleUrl) {
|
|
|
367
367
|
}
|
|
368
368
|
/** The ambient build-time version, or null outside a tsdown build. */
|
|
369
369
|
function embeddedVersion() {
|
|
370
|
-
return "1.6.3-beta.
|
|
370
|
+
return "1.6.3-beta.15";
|
|
371
371
|
}
|
|
372
372
|
/**
|
|
373
373
|
* Resolve the running Pantheon version for the what's-new window: the plugin
|
|
@@ -3752,7 +3752,7 @@ function Sidebar(props) {
|
|
|
3752
3752
|
setProp(_el$12, "width", "100%");
|
|
3753
3753
|
setProp(_el$12, "padding", 1);
|
|
3754
3754
|
setProp(_el$13, "attributes", 1);
|
|
3755
|
-
insert(_el$13, `Pantheon v1.6.3-beta.
|
|
3755
|
+
insert(_el$13, `Pantheon v1.6.3-beta.15`);
|
|
3756
3756
|
insert(_el$14, () => presetHeaderLine(model.preset().name ?? "", model.provider()));
|
|
3757
3757
|
insert(_el$12, createComponent(Show, {
|
|
3758
3758
|
get when() {
|
|
@@ -4473,7 +4473,7 @@ function HomeStatus(props) {
|
|
|
4473
4473
|
const muted = () => props.context.theme?.text?.muted ?? props.context.theme?.text?.base;
|
|
4474
4474
|
return (() => {
|
|
4475
4475
|
var _el$72 = createElement("text");
|
|
4476
|
-
insert(_el$72, () => `Pantheon v1.6.3-beta.
|
|
4476
|
+
insert(_el$72, () => `Pantheon v1.6.3-beta.15${preset().name ? ` ▸ ${preset().name}` : ""}`);
|
|
4477
4477
|
effect((_$p) => setProp(_el$72, "fg", muted(), _$p));
|
|
4478
4478
|
return _el$72;
|
|
4479
4479
|
})();
|
|
@@ -1,12 +1,12 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "pantheon-tui",
|
|
3
|
-
"version": "1.6.3-beta.
|
|
3
|
+
"version": "1.6.3-beta.17",
|
|
4
4
|
"lockfileVersion": 3,
|
|
5
5
|
"requires": true,
|
|
6
6
|
"packages": {
|
|
7
7
|
"": {
|
|
8
8
|
"name": "pantheon-tui",
|
|
9
|
-
"version": "1.6.3-beta.
|
|
9
|
+
"version": "1.6.3-beta.17",
|
|
10
10
|
"license": "MIT",
|
|
11
11
|
"dependencies": {
|
|
12
12
|
"@opencode-ai/plugin": "1.18.33",
|
|
@@ -1,7 +1,7 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "pantheon-tui",
|
|
3
3
|
"type": "module",
|
|
4
|
-
"version": "1.6.3-beta.
|
|
4
|
+
"version": "1.6.3-beta.17",
|
|
5
5
|
"description": "Pantheon TUI plugin — V1 rich sidebar and usage gauges; native V2 status, session and delegation sidebar",
|
|
6
6
|
"license": "MIT",
|
|
7
7
|
"engines": {
|
package/src/routing.yml
CHANGED
package/src/skills/README.md
CHANGED
|
@@ -4,7 +4,7 @@
|
|
|
4
4
|
|
|
5
5
|
Skills are reference modules loaded on demand by agents. Each skill provides domain-specific knowledge — architecture patterns, security checklists, optimization strategies, and implementation guides. Agents load skills when their task matches the skill's description, keeping context focused and token-efficient.
|
|
6
6
|
|
|
7
|
-
There are **
|
|
7
|
+
There are **21 skills** divided into **5 domains**.
|
|
8
8
|
|
|
9
9
|
## Skills by Domain
|
|
10
10
|
|
|
@@ -0,0 +1,41 @@
|
|
|
1
|
+
---
|
|
2
|
+
name: prompt-refinery
|
|
3
|
+
description: "Turn a rough, vague, or one-line request into a structured execution contract (objective, scope slices, acceptance criteria, deliverables, open questions) shown to the user BEFORE executing. Use when a request spans multiple concerns or lacks explicit acceptance criteria."
|
|
4
|
+
---
|
|
5
|
+
|
|
6
|
+
# Prompt Refinery
|
|
7
|
+
|
|
8
|
+
> **A vague request executes expensively and wrong. Refine it into a contract, show the contract, then execute what was confirmed.**
|
|
9
|
+
|
|
10
|
+
---
|
|
11
|
+
|
|
12
|
+
## When this fires
|
|
13
|
+
|
|
14
|
+
- The request spans multiple concerns ("refatora o dashboard, corrige os cálculos, melhora a UI") with no explicit acceptance criteria.
|
|
15
|
+
- The request names a desired OUTCOME but not the slices, nomenclature, data sources, or quality bar.
|
|
16
|
+
- The user pastes a long context dump and asks for "melhorar/refatorar/revisar" without structure.
|
|
17
|
+
|
|
18
|
+
A one-line bounded fix ("troque essa cor", "corrija esse typo") does NOT fire this skill — execute directly.
|
|
19
|
+
|
|
20
|
+
---
|
|
21
|
+
|
|
22
|
+
## The contract shape (the rewrite target)
|
|
23
|
+
|
|
24
|
+
Produce ALL of the following sections, filling from the request and marking real gaps as open questions — never invent requirements silently:
|
|
25
|
+
|
|
26
|
+
1. **Objetivo do projeto** — one paragraph: what changes and why.
|
|
27
|
+
2. **Escopo em fatias numeradas** — each slice independently verifiable (audit/corrections; architecture/IA; nomenclature/statistics; UI/UX; …). Slices the user did not ask for go under "fora de escopo", not here.
|
|
28
|
+
3. **Critérios de aceite por fatia** — testable statements ("toda moeda passa por Decimal; nenhum float em valor monetário"), not vibes.
|
|
29
|
+
4. **Entregáveis esperados** — the concrete artifacts (report, route map, dictionary, implementation).
|
|
30
|
+
5. **Perguntas abertas** — the gaps that change the plan, with your recommended default for each so silence means "proceed with the default".
|
|
31
|
+
6. **Glossário/nomenclatura proposta** — when the domain has ambiguous terms, propose the canonical names up front.
|
|
32
|
+
|
|
33
|
+
---
|
|
34
|
+
|
|
35
|
+
## Operating rules
|
|
36
|
+
|
|
37
|
+
- **Show, then execute.** The refined contract is presented to the user for confirm-or-adjust BEFORE implementation — one confirmation, not a dialogue. Silence or "ok" proceeds with the defaults stated in Perguntas abertas.
|
|
38
|
+
- **Domain mapping stays the ladder's**: after confirmation, slices route through the normal discovery ladder (local selector → Jev → specialist dispatch). The contract is the SPEC, not the dispatcher.
|
|
39
|
+
- **Preserve the user's language** (a PT-BR request refines into PT-BR).
|
|
40
|
+
- **One refinement turn.** The rewrite is a single planning pass on the session model — no extra provider call, no Jev needed for the rewrite itself (decision models answer typed scalars, they do not write prose).
|
|
41
|
+
- **Cost honesty**: the refinement costs roughly one planning turn (~1–4K tokens each way). The expensive thing is executing a mis-specified request; this skill buys that back.
|