pantheon-opencode 1.6.1 → 1.6.2-beta.10
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/AGENTS.global.md +11 -11
- package/AGENTS.md +2 -2
- package/CHANGELOG.md +179 -0
- package/README.md +268 -20
- package/README.pt-BR.md +34 -11
- package/bin/pantheon-init.mjs +301 -32
- package/package.json +13 -4
- package/plugin.json +1 -1
- package/prompts/README.md +58 -0
- package/prompts/debug-issue.prompt.md +46 -0
- package/prompts/dynamic/agora-generated.txt +38 -0
- package/prompts/dynamic/orchestrate-generated.txt +51 -0
- package/prompts/focus.prompt.md +42 -0
- package/prompts/implement-feature.prompt.md +39 -0
- package/prompts/optimize-database.prompt.md +58 -0
- package/prompts/orchestrate-with-zeus.prompt.md +441 -0
- package/prompts/plan-architecture.prompt.md +59 -0
- package/prompts/quick-discovery-large-codebase.prompt.md +294 -0
- package/prompts/quick-plan-large-feature.prompt.md +197 -0
- package/prompts/semantic-summarize.md +50 -0
- package/prompts/sketch.prompt.md +91 -0
- package/prompts/subtask.prompt.md +131 -0
- package/pyproject.toml +1 -1
- package/scripts/build-agents-md.mjs +1 -1
- package/scripts/decision-router-budget.mjs +780 -0
- package/scripts/decision-router-v2-budget.mjs +218 -0
- package/scripts/decision-router-v2-egress.mjs +272 -0
- package/scripts/decision-router.mjs +758 -0
- package/scripts/docs-check-changelog.mjs +5 -3
- package/scripts/doctor.mjs +213 -45
- package/scripts/install/agents-md.mjs +2 -1
- package/scripts/install/component-selection.mjs +9 -4
- package/scripts/install/config-migration.mjs +344 -393
- package/scripts/install/config-paths.mjs +15 -0
- package/scripts/install/decision-layer-preview.mjs +376 -0
- package/scripts/install/decision-router-config.mjs +560 -0
- package/scripts/install/health-check.mjs +20 -11
- package/scripts/install/interactive.mjs +103 -14
- package/scripts/install/interrupt.mjs +4 -3
- package/scripts/install/mcp-permissions.mjs +164 -0
- package/scripts/install/migrate.mjs +123 -72
- package/scripts/install/opencode-version.mjs +31 -13
- package/scripts/install/opencode-wildcard.mjs +106 -0
- package/scripts/install/opencode.mjs +411 -184
- package/scripts/install/plan-artifact.mjs +12 -1
- package/scripts/install/plan.mjs +8 -1
- package/scripts/install/plugin.mjs +18 -1
- package/scripts/install/state.mjs +12 -0
- package/scripts/install/strings.mjs +22 -0
- package/scripts/install/v2-capabilities.mjs +5 -6
- package/scripts/install/venv.mjs +34 -17
- package/scripts/mcp_resources.py +11 -9
- package/scripts/package-evidence.mjs +1 -0
- package/scripts/release-notes.mjs +30 -7
- package/scripts/restore.mjs +22 -10
- package/scripts/sandbox-env.sh +174 -0
- package/scripts/setup-v2-config.mjs +1703 -0
- package/scripts/setup.mjs +415 -0
- package/scripts/sync-tui.mjs +9 -1
- package/scripts/test-opencode-v2-sandbox.sh +448 -157
- package/scripts/update-skills-lock.mjs +30 -0
- package/scripts/versioning.mjs +41 -5
- package/skills-lock.json +10 -10
- package/src/agents/aphrodite.md +11 -3
- package/src/agents/apollo.md +6 -0
- package/src/agents/athena.md +7 -1
- package/src/agents/demeter.md +6 -2
- package/src/agents/gaia.md +5 -0
- package/src/agents/hephaestus.md +7 -2
- package/src/agents/hermes.md +14 -13
- package/src/agents/iris.md +6 -2
- package/src/agents/mnemosyne.md +21 -50
- package/src/agents/nyx.md +6 -2
- package/src/agents/prometheus.md +5 -1
- package/src/agents/talos.md +10 -1
- package/src/agents/themis.md +5 -1
- package/src/agents/zeus.md +61 -65
- package/src/instructions/agent-return-format.instructions.md +1 -1
- package/src/instructions/memory-protocol.instructions.md +3 -3
- package/src/instructions/test-economy.instructions.md +6 -6
- package/src/mcp/eval_store.py +11 -4
- package/src/mcp/mcp_resources.py +11 -9
- package/src/mcp/memory_mcp.py +293 -150
- package/src/mcp/pantheon_vision.py +37 -36
- package/src/pantheon/cost-command.ts +30 -133
- package/src/pantheon/jev-decision-tool.ts +347 -0
- package/src/pantheon/jev-decisions.ts +625 -0
- package/src/pantheon/local-decision-agents.ts +170 -0
- package/src/pantheon/local-decision-budget.ts +1046 -0
- package/src/pantheon/local-decision-egress-gate.ts +144 -0
- package/src/pantheon/local-decision-prompt.ts +256 -0
- package/src/pantheon/local-decision-provider-matrix.ts +44 -0
- package/src/pantheon/local-decision-router.ts +225 -0
- package/src/pantheon/local-decision-skills.ts +281 -0
- package/src/pantheon/local-decision-tool.ts +160 -0
- package/src/pantheon/local-decision-turn.ts +101 -0
- package/src/pantheon/local-decision-usage-evidence.internal.ts +168 -0
- package/src/pantheon/local-orchestration.ts +322 -0
- package/src/pantheon/model-pricing.ts +1 -536
- package/src/pantheon/pantheon-decision-catalog.ts +101 -0
- package/src/pantheon/v2-decision-budget.ts +456 -0
- package/src/pantheon/v2-decision-contracts.ts +300 -0
- package/src/pantheon/v2-decision-egress.ts +203 -0
- package/src/pantheon/v2-decision-events.ts +180 -0
- package/src/pantheon/v2-decision-heuristic.ts +121 -0
- package/src/pantheon/v2-decision-points/p1-agent.ts +91 -0
- package/src/pantheon/v2-decision-points/p10-claims.ts +243 -0
- package/src/pantheon/v2-decision-points/p11-catalog.ts +220 -0
- package/src/pantheon/v2-decision-points/p2-skill.ts +83 -0
- package/src/pantheon/v2-decision-points/p3-complexity.ts +108 -0
- package/src/pantheon/v2-decision-points/p4-memory.ts +137 -0
- package/src/pantheon/v2-decision-points/p5-themis-verify.ts +190 -0
- package/src/pantheon/v2-decision-points/p6-council.ts +161 -0
- package/src/pantheon/v2-decision-points/p7-loop.ts +263 -0
- package/src/pantheon/v2-decision-points/p8-guardrail.ts +367 -0
- package/src/pantheon/v2-decision-points/p9-memory-compaction.ts +304 -0
- package/src/pantheon/v2-decision-points.ts +242 -0
- package/src/pantheon/v2-decision-policy.ts +166 -0
- package/src/pantheon/v2-decision-premises.ts +146 -0
- package/src/pantheon/v2-decision-prompt-surface.ts +115 -0
- package/src/pantheon/v2-decision-rank.ts +91 -0
- package/src/pantheon/v2-decision-registry.ts +391 -0
- package/src/pantheon/v2-decision-resource-ids.ts +127 -0
- package/src/pantheon/v2-decision-shadow.ts +123 -0
- package/src/pantheon/v2-decision-telemetry.ts +214 -0
- package/src/pantheon/v2-decision-verify.ts +51 -0
- package/src/pantheon/v2-hook-engine.ts +307 -0
- package/src/pantheon/v2-native-decisions.ts +365 -0
- package/src/pantheon/v2-shell-hooks.ts +66 -2
- package/src/pantheon/v2-tools.ts +38 -15
- package/src/pantheon/v2-vision.ts +34 -29
- package/src/pantheon/vision.ts +43 -10
- package/src/plugin-v2.ts +449 -79
- package/src/plugin.ts +4 -0
- package/src/plugins/tui/THIRD_PARTY_NOTICES.md +24 -0
- package/src/plugins/tui/dist/server-v2.d.ts +22 -0
- package/src/plugins/tui/dist/server-v2.d.ts.map +1 -0
- package/src/plugins/tui/dist/server-v2.js +26 -0
- package/src/plugins/tui/dist/server-v2.js.map +1 -0
- package/src/plugins/tui/dist/tui-v2.d.ts +54 -51
- package/src/plugins/tui/dist/tui-v2.d.ts.map +1 -1
- package/src/plugins/tui/dist/tui-v2.js +762 -529
- package/src/plugins/tui/dist/tui-v2.js.map +1 -1
- package/src/plugins/tui/dist/tui.js +1 -1
- package/src/plugins/tui/index-v2.ts +3 -6
- package/src/plugins/tui/package-lock.json +17 -2
- package/src/plugins/tui/package.json +6 -3
- package/src/plugins/tui/src/context-gauge.ts +59 -0
- package/src/plugins/tui/src/cost-data.ts +607 -0
- package/src/plugins/tui/src/index-v2.tsx +360 -242
- package/src/plugins/tui/src/model-pricing.ts +536 -0
- package/src/plugins/tui/src/model-usage-limits.ts +4 -0
- package/src/plugins/tui/src/server-v2.ts +21 -0
- package/src/plugins/tui/src/sidebar-width.ts +129 -0
- package/src/plugins/tui/tsconfig.typecheck.json +1 -1
- package/src/plugins/tui/tsdown.config.ts +34 -5
- package/src/routing.yml +72 -13
- package/src/skills/README.md +4 -0
- package/src/skills/agent-coordination/SKILL.md +0 -3
- package/src/skills/artifact-management/SKILL.md +0 -3
- package/src/skills/auto-continue/SKILL.md +0 -3
- package/src/skills/code-review-checklist/SKILL.md +0 -3
- package/src/skills/context-compression/SKILL.md +0 -3
- package/src/skills/memory-bank/SKILL.md +6 -28
- package/src/skills/session-goal/SKILL.md +0 -3
- package/src/skills/simplify/SKILL.md +5 -4
- package/src/skills/tdd-with-agents/SKILL.md +0 -3
- package/src/skills/visual-review-pipeline/SKILL.md +0 -1
- package/src/pantheon/memory-context.ts +0 -1017
|
@@ -0,0 +1,214 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* V2 decision telemetry — allowlist sanitisation, closing defect D10.
|
|
3
|
+
*
|
|
4
|
+
* Telemetry must never carry prompt, response, user content, credentials or
|
|
5
|
+
* free-form text. The strongest way to guarantee that is not to scrub known
|
|
6
|
+
* keys but to accept only a small allowlist whose values are constrained to
|
|
7
|
+
* short tokens and bounded numbers:
|
|
8
|
+
*
|
|
9
|
+
* schemaVersion, point, decision, effect, reason, mode, durationMs, candidateCount
|
|
10
|
+
*
|
|
11
|
+
* Any other key is dropped and only a count is reported. Neither the key nor
|
|
12
|
+
* its value is ever copied into the result, so no attacker-controlled string
|
|
13
|
+
* reaches telemetry through a rejection path.
|
|
14
|
+
*
|
|
15
|
+
* `effect` is fixed to `'none'`. An event that claims anything else is rejected
|
|
16
|
+
* outright: this layer has no measured effect to report, and silently rewriting
|
|
17
|
+
* the claim would hide the caller's error.
|
|
18
|
+
*
|
|
19
|
+
* The sink is in-memory and bounded. It performs no I/O, so the offline harness
|
|
20
|
+
* can assert exactly what would be emitted.
|
|
21
|
+
*/
|
|
22
|
+
|
|
23
|
+
import type { DecisionPointId } from './v2-decision-registry.ts'
|
|
24
|
+
|
|
25
|
+
export const TELEMETRY_SCHEMA_VERSION = 1
|
|
26
|
+
export const TELEMETRY_FIELDS = [
|
|
27
|
+
'schemaVersion',
|
|
28
|
+
'point',
|
|
29
|
+
'decision',
|
|
30
|
+
'effect',
|
|
31
|
+
'reason',
|
|
32
|
+
'mode',
|
|
33
|
+
'durationMs',
|
|
34
|
+
'candidateCount',
|
|
35
|
+
] as const
|
|
36
|
+
|
|
37
|
+
export type TelemetryField = (typeof TELEMETRY_FIELDS)[number]
|
|
38
|
+
export type TelemetryMode = 'shadow' | 'active' | 'off' | 'migration'
|
|
39
|
+
|
|
40
|
+
export interface DecisionTelemetryEvent {
|
|
41
|
+
readonly schemaVersion: typeof TELEMETRY_SCHEMA_VERSION
|
|
42
|
+
readonly point: DecisionPointId
|
|
43
|
+
readonly decision: string
|
|
44
|
+
readonly effect: 'none'
|
|
45
|
+
readonly reason: string
|
|
46
|
+
readonly mode?: TelemetryMode
|
|
47
|
+
readonly durationMs?: number
|
|
48
|
+
readonly candidateCount?: number
|
|
49
|
+
}
|
|
50
|
+
|
|
51
|
+
export type TelemetryRejection = 'INVALID_INPUT' | 'MISSING_REQUIRED_FIELD' | 'EFFECT_NOT_MEASURED'
|
|
52
|
+
|
|
53
|
+
export type TelemetryResult =
|
|
54
|
+
| {
|
|
55
|
+
readonly ok: true
|
|
56
|
+
readonly event: DecisionTelemetryEvent
|
|
57
|
+
readonly droppedFieldCount: number
|
|
58
|
+
}
|
|
59
|
+
| {
|
|
60
|
+
readonly ok: false
|
|
61
|
+
readonly reasonCode: TelemetryRejection
|
|
62
|
+
readonly droppedFieldCount: number
|
|
63
|
+
}
|
|
64
|
+
|
|
65
|
+
const TOKEN = /^[A-Za-z0-9][A-Za-z0-9._:-]{0,63}$/
|
|
66
|
+
const POINT_ID = /^P(?:[1-9]|1[01])$/
|
|
67
|
+
const TELEMETRY_KEYS: ReadonlySet<string> = new Set(TELEMETRY_FIELDS)
|
|
68
|
+
const MODES: ReadonlySet<string> = new Set(['shadow', 'active', 'off', 'migration'])
|
|
69
|
+
const DURATION_MAX_MS = 60_000
|
|
70
|
+
const CANDIDATE_MAX = 10_000
|
|
71
|
+
function isRecord(value: unknown): value is Record<string, unknown> {
|
|
72
|
+
return typeof value === 'object' && value !== null && !Array.isArray(value)
|
|
73
|
+
}
|
|
74
|
+
|
|
75
|
+
function isToken(value: unknown): value is string {
|
|
76
|
+
return typeof value === 'string' && TOKEN.test(value)
|
|
77
|
+
}
|
|
78
|
+
|
|
79
|
+
function isPointId(value: unknown): value is DecisionPointId {
|
|
80
|
+
return typeof value === 'string' && POINT_ID.test(value)
|
|
81
|
+
}
|
|
82
|
+
|
|
83
|
+
function rejected(reasonCode: TelemetryRejection, droppedFieldCount: number): TelemetryResult {
|
|
84
|
+
return { ok: false, reasonCode, droppedFieldCount }
|
|
85
|
+
}
|
|
86
|
+
|
|
87
|
+
/**
|
|
88
|
+
* Reduce an arbitrary record to the telemetry allowlist.
|
|
89
|
+
*
|
|
90
|
+
* Required survivors: `point`, `decision` and `reason`. `effect`, when present,
|
|
91
|
+
* must already be `'none'`; the output always carries `'none'`.
|
|
92
|
+
*/
|
|
93
|
+
export function sanitizeTelemetry(input: unknown): TelemetryResult {
|
|
94
|
+
if (!isRecord(input)) return rejected('INVALID_INPUT', 0)
|
|
95
|
+
|
|
96
|
+
let droppedFieldCount = 0
|
|
97
|
+
const drop = (): void => {
|
|
98
|
+
droppedFieldCount += 1
|
|
99
|
+
}
|
|
100
|
+
|
|
101
|
+
let point: DecisionPointId | null = null
|
|
102
|
+
let decision: string | null = null
|
|
103
|
+
let reason: string | null = null
|
|
104
|
+
let mode: TelemetryMode | null = null
|
|
105
|
+
let durationMs: number | null = null
|
|
106
|
+
let candidateCount: number | null = null
|
|
107
|
+
let effectNotMeasured = false
|
|
108
|
+
|
|
109
|
+
for (const key of Object.keys(input).sort()) {
|
|
110
|
+
const value = input[key]
|
|
111
|
+
if (!TELEMETRY_KEYS.has(key)) {
|
|
112
|
+
drop()
|
|
113
|
+
continue
|
|
114
|
+
}
|
|
115
|
+
switch (key) {
|
|
116
|
+
case 'schemaVersion':
|
|
117
|
+
if (value !== TELEMETRY_SCHEMA_VERSION) drop()
|
|
118
|
+
break
|
|
119
|
+
case 'point':
|
|
120
|
+
if (isPointId(value)) point = value
|
|
121
|
+
else drop()
|
|
122
|
+
break
|
|
123
|
+
case 'decision':
|
|
124
|
+
if (isToken(value)) decision = value
|
|
125
|
+
else drop()
|
|
126
|
+
break
|
|
127
|
+
case 'reason':
|
|
128
|
+
if (isToken(value)) reason = value
|
|
129
|
+
else drop()
|
|
130
|
+
break
|
|
131
|
+
case 'mode':
|
|
132
|
+
if (typeof value === 'string' && MODES.has(value)) {
|
|
133
|
+
mode = value as TelemetryMode
|
|
134
|
+
} else {
|
|
135
|
+
drop()
|
|
136
|
+
}
|
|
137
|
+
break
|
|
138
|
+
case 'durationMs':
|
|
139
|
+
if (
|
|
140
|
+
typeof value === 'number' &&
|
|
141
|
+
Number.isFinite(value) &&
|
|
142
|
+
value >= 0 &&
|
|
143
|
+
value <= DURATION_MAX_MS
|
|
144
|
+
) {
|
|
145
|
+
durationMs = value
|
|
146
|
+
} else {
|
|
147
|
+
drop()
|
|
148
|
+
}
|
|
149
|
+
break
|
|
150
|
+
case 'candidateCount':
|
|
151
|
+
if (
|
|
152
|
+
Number.isInteger(value) &&
|
|
153
|
+
(value as number) >= 0 &&
|
|
154
|
+
(value as number) <= CANDIDATE_MAX
|
|
155
|
+
) {
|
|
156
|
+
candidateCount = value as number
|
|
157
|
+
} else {
|
|
158
|
+
drop()
|
|
159
|
+
}
|
|
160
|
+
break
|
|
161
|
+
case 'effect':
|
|
162
|
+
if (value === 'none') break
|
|
163
|
+
effectNotMeasured = true
|
|
164
|
+
drop()
|
|
165
|
+
break
|
|
166
|
+
default:
|
|
167
|
+
drop()
|
|
168
|
+
}
|
|
169
|
+
}
|
|
170
|
+
|
|
171
|
+
if (effectNotMeasured) return rejected('EFFECT_NOT_MEASURED', droppedFieldCount)
|
|
172
|
+
if (point === null || decision === null || reason === null) {
|
|
173
|
+
return rejected('MISSING_REQUIRED_FIELD', droppedFieldCount)
|
|
174
|
+
}
|
|
175
|
+
|
|
176
|
+
const event: DecisionTelemetryEvent = {
|
|
177
|
+
schemaVersion: TELEMETRY_SCHEMA_VERSION,
|
|
178
|
+
point,
|
|
179
|
+
decision,
|
|
180
|
+
effect: 'none',
|
|
181
|
+
reason,
|
|
182
|
+
...(mode !== null ? { mode } : {}),
|
|
183
|
+
...(durationMs !== null ? { durationMs } : {}),
|
|
184
|
+
...(candidateCount !== null ? { candidateCount } : {}),
|
|
185
|
+
}
|
|
186
|
+
return { ok: true, event: Object.freeze(event), droppedFieldCount }
|
|
187
|
+
}
|
|
188
|
+
|
|
189
|
+
/** In-memory, bounded telemetry collector. No I/O, no network. */
|
|
190
|
+
export interface TelemetrySink {
|
|
191
|
+
emit(input: unknown): TelemetryResult
|
|
192
|
+
events(): readonly DecisionTelemetryEvent[]
|
|
193
|
+
size(): number
|
|
194
|
+
}
|
|
195
|
+
|
|
196
|
+
export function createTelemetrySink(maxEvents = 1_000): TelemetrySink {
|
|
197
|
+
if (!Number.isInteger(maxEvents) || maxEvents < 0 || maxEvents > 100_000) {
|
|
198
|
+
throw new RangeError('maxEvents must be an integer in 0..100000')
|
|
199
|
+
}
|
|
200
|
+
const collected: DecisionTelemetryEvent[] = []
|
|
201
|
+
return {
|
|
202
|
+
emit(input: unknown) {
|
|
203
|
+
const result = sanitizeTelemetry(input)
|
|
204
|
+
if (result.ok && collected.length < maxEvents) collected.push(result.event)
|
|
205
|
+
return result
|
|
206
|
+
},
|
|
207
|
+
events() {
|
|
208
|
+
return Object.freeze([...collected])
|
|
209
|
+
},
|
|
210
|
+
size() {
|
|
211
|
+
return collected.length
|
|
212
|
+
},
|
|
213
|
+
}
|
|
214
|
+
}
|
|
@@ -0,0 +1,51 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* V2 verification contract — explicit verdicts, never inferred.
|
|
3
|
+
*
|
|
4
|
+
* `failed_to_respond` during mandatory verification is `BLOCK`; during advisory
|
|
5
|
+
* verification it is `FAIL`. Neither is ever `PASS`, and no scalar can stand in
|
|
6
|
+
* for a verdict.
|
|
7
|
+
*/
|
|
8
|
+
|
|
9
|
+
export const VERIFY_VERDICTS = ['PASS', 'FAIL', 'BLOCK'] as const
|
|
10
|
+
export type VerifyVerdict = (typeof VERIFY_VERDICTS)[number]
|
|
11
|
+
|
|
12
|
+
export interface VerifyRequirement {
|
|
13
|
+
readonly id: string
|
|
14
|
+
readonly mandatory: boolean
|
|
15
|
+
}
|
|
16
|
+
|
|
17
|
+
export type VerifyReasonCode = 'VERDICT' | 'FAILED_TO_RESPOND' | 'UNRECOGNIZED_VERDICT'
|
|
18
|
+
|
|
19
|
+
export interface VerifyOutcome {
|
|
20
|
+
readonly requirementId: string
|
|
21
|
+
readonly verdict: VerifyVerdict
|
|
22
|
+
readonly reasonCode: VerifyReasonCode
|
|
23
|
+
readonly mandatory: boolean
|
|
24
|
+
}
|
|
25
|
+
|
|
26
|
+
/**
|
|
27
|
+
* Resolve a verification verdict.
|
|
28
|
+
*
|
|
29
|
+
* The verdict must be stated explicitly. `failed_to_respond` during a
|
|
30
|
+
* mandatory verification is `BLOCK`; during an advisory verification it is
|
|
31
|
+
* `FAIL`. Neither case is ever `PASS`, and no certainty scalar can stand in
|
|
32
|
+
* for a verdict.
|
|
33
|
+
*/
|
|
34
|
+
export function resolveVerify(requirement: VerifyRequirement, response: unknown): VerifyOutcome {
|
|
35
|
+
const base = { requirementId: requirement.id, mandatory: requirement.mandatory }
|
|
36
|
+
if (typeof response === 'string') {
|
|
37
|
+
if (response === 'PASS' || response === 'FAIL' || response === 'BLOCK') {
|
|
38
|
+
return { ...base, verdict: response, reasonCode: 'VERDICT' }
|
|
39
|
+
}
|
|
40
|
+
return {
|
|
41
|
+
...base,
|
|
42
|
+
verdict: requirement.mandatory ? 'BLOCK' : 'FAIL',
|
|
43
|
+
reasonCode: 'UNRECOGNIZED_VERDICT',
|
|
44
|
+
}
|
|
45
|
+
}
|
|
46
|
+
return {
|
|
47
|
+
...base,
|
|
48
|
+
verdict: requirement.mandatory ? 'BLOCK' : 'FAIL',
|
|
49
|
+
reasonCode: 'FAILED_TO_RESPOND',
|
|
50
|
+
}
|
|
51
|
+
}
|
|
@@ -0,0 +1,307 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* In-process replacements for the three bash hooks the V2 tool guards used to
|
|
3
|
+
* spawn per tool call (`scan-secrets.sh`, `on-subagent-delegation-start.sh`,
|
|
4
|
+
* `on-subagent-delegation-stop.sh`).
|
|
5
|
+
*
|
|
6
|
+
* Why: the shell path cost one `bash` spawn per tool call plus two more per
|
|
7
|
+
* delegation (each of those also spawning `python3` to derive JSON fields), on
|
|
8
|
+
* the hot path of every tool execution. The logic is pure string processing —
|
|
9
|
+
* pattern matching and log-line formatting — so it runs in this process with
|
|
10
|
+
* identical observable behavior: same exit-code contract (0 allow / 1 advisory
|
|
11
|
+
* / 2 block), same masked `[SECRET SCAN]` trail, same `delegations.log`
|
|
12
|
+
* JSON lines, same LOG_DIR resolution order.
|
|
13
|
+
*
|
|
14
|
+
* What deliberately does NOT move here: `format-multi-language.sh` still
|
|
15
|
+
* spawns when it can actually format (it shells out to formatters itself);
|
|
16
|
+
* the in-process gate only skips the spawn for tool calls the script would
|
|
17
|
+
* no-op anyway. `HOOK_TIMEOUT_MS` does not apply — these functions are
|
|
18
|
+
* synchronous and bounded by the payload size, not by a child process.
|
|
19
|
+
*
|
|
20
|
+
* Fail-closed parity: an in-process scanner has no "missing script" failure,
|
|
21
|
+
* but an unexpected internal error still refuses to produce a trustworthy
|
|
22
|
+
* verdict — it returns exit-code 1 WITHOUT the `[SECRET SCAN]` marker, which
|
|
23
|
+
* the existing classifier (plugin-v2 `evaluateSecretScanResult`) reads as
|
|
24
|
+
* fail-closed, exactly as a broken child path always did.
|
|
25
|
+
*
|
|
26
|
+
* @module pantheon/v2-hook-engine
|
|
27
|
+
*/
|
|
28
|
+
import { appendFileSync, mkdirSync } from 'node:fs'
|
|
29
|
+
import { join } from 'node:path'
|
|
30
|
+
import type { HookPayload, HookResult } from '../plugins/hook-runner.ts'
|
|
31
|
+
|
|
32
|
+
/** Mirrors the scanner trail marker the classifier keys the advisory on. */
|
|
33
|
+
export const SECRET_SCAN_MARKER = '[SECRET SCAN]'
|
|
34
|
+
|
|
35
|
+
// Provider-specific markers are assembled at runtime so this scanner is not
|
|
36
|
+
// self-matched by the repository secret scan (the shell script's own trick).
|
|
37
|
+
const BIFROST_TOKEN_PREFIX = `${'sk'}${'-bf-'}`
|
|
38
|
+
const BIFROST_HEADER_NAME = `${'x'}${'-bf-'}${'vk'}`
|
|
39
|
+
|
|
40
|
+
/**
|
|
41
|
+
* High-confidence token formats. MIRRORS `HIGH_CONFIDENCE_PATTERNS` in
|
|
42
|
+
* scripts/hooks/scan-secrets.sh and `SECRET_MASK_RE` in src/plugin-v2.ts
|
|
43
|
+
* (V2 log masking); the drift is asserted by tests.
|
|
44
|
+
*/
|
|
45
|
+
const HIGH_CONFIDENCE_RES: readonly RegExp[] = Object.freeze([
|
|
46
|
+
/AKIA[0-9A-Z]{16}/i,
|
|
47
|
+
/gh[pousr]_[A-Za-z0-9_]{36,}/i,
|
|
48
|
+
/glpat-[A-Za-z0-9_-]{20}/i,
|
|
49
|
+
/sk-[a-zA-Z0-9]{20,}/i,
|
|
50
|
+
/sk_live_[a-zA-Z0-9]{20,}/i,
|
|
51
|
+
/sk_test_[a-zA-Z0-9]{20,}/i,
|
|
52
|
+
/xox[baprs]-[0-9]{10,13}-[0-9]{10,13}-[a-zA-Z0-9]{24}/i,
|
|
53
|
+
/bearer\s+[a-zA-Z0-9_.-]{20,}/i,
|
|
54
|
+
/eyJ[A-Za-z0-9_-]*\.eyJ[A-Za-z0-9_-]*\.[A-Za-z0-9_-]/,
|
|
55
|
+
])
|
|
56
|
+
const BIFROST_TOKEN_RE = new RegExp(`${BIFROST_TOKEN_PREFIX}[A-Za-z0-9_-]{8,}`, 'i')
|
|
57
|
+
|
|
58
|
+
/**
|
|
59
|
+
* KEY/header names without a recognizable token value — advisory only, never
|
|
60
|
+
* blocks (the script's LOW_CONFIDENCE_PATTERNS, including the Bifrost header).
|
|
61
|
+
*/
|
|
62
|
+
const LOW_CONFIDENCE_RES: readonly RegExp[] = Object.freeze([
|
|
63
|
+
/[a-zA-Z0-9_-]*api[_-]?key[a-zA-Z0-9_-]*\s*[:=]\s*["']?[a-zA-Z0-9_-]{16,}["']?/,
|
|
64
|
+
/[a-zA-Z0-9_-]*password[a-zA-Z0-9_-]*\s*[:=]\s*["'][^"']{8,}["']/,
|
|
65
|
+
/[a-zA-Z0-9_-]*secret[a-zA-Z0-9_-]*\s*[:=]\s*["'][^"']{8,}["']/,
|
|
66
|
+
])
|
|
67
|
+
const BIFROST_HEADER_RE = new RegExp(BIFROST_HEADER_NAME, 'i')
|
|
68
|
+
|
|
69
|
+
/** First 4 + last 4 chars (len > 8), else '****' — mirrors `mask_secret`. */
|
|
70
|
+
export function maskSecretMatch(match: string): string {
|
|
71
|
+
return match.length <= 8 ? '****' : `${match.slice(0, 4)}****${match.slice(-4)}`
|
|
72
|
+
}
|
|
73
|
+
|
|
74
|
+
function firstMatch(input: string, pattern: RegExp): string | null {
|
|
75
|
+
const found = input.match(pattern)
|
|
76
|
+
return found === null ? null : found[0]
|
|
77
|
+
}
|
|
78
|
+
|
|
79
|
+
/**
|
|
80
|
+
* Known read-only tools: their INPUT cannot carry a secret out of the user's
|
|
81
|
+
* machine (paths, patterns, ids), so scanning them was pure latency. The gate
|
|
82
|
+
* is a DENYLIST — any tool this list has never heard of keeps getting scanned
|
|
83
|
+
* (fail-closed for unknown names).
|
|
84
|
+
*/
|
|
85
|
+
const READ_ONLY_TOOLS = new Set([
|
|
86
|
+
'read',
|
|
87
|
+
'grep',
|
|
88
|
+
'glob',
|
|
89
|
+
'list',
|
|
90
|
+
'question',
|
|
91
|
+
'session_list',
|
|
92
|
+
'session_read',
|
|
93
|
+
'worktree_list',
|
|
94
|
+
'subagent_list',
|
|
95
|
+
])
|
|
96
|
+
|
|
97
|
+
export function secretScanApplies(tool: unknown): boolean {
|
|
98
|
+
if (typeof tool !== 'string') return true
|
|
99
|
+
const normalized = tool.trim().toLowerCase()
|
|
100
|
+
if (normalized === '') return true
|
|
101
|
+
return !READ_ONLY_TOOLS.has(normalized)
|
|
102
|
+
}
|
|
103
|
+
|
|
104
|
+
/** The serialized tool input the shell scanner received as `$1`/stdin. */
|
|
105
|
+
function scannerInput(payload: HookPayload): string {
|
|
106
|
+
if (typeof payload === 'string') return payload
|
|
107
|
+
if (payload === null || payload === undefined) return ''
|
|
108
|
+
const record = payload as Record<string, unknown>
|
|
109
|
+
const input = record.tool_input ?? record.input ?? payload
|
|
110
|
+
if (typeof input === 'string') return input
|
|
111
|
+
try {
|
|
112
|
+
return JSON.stringify(input)
|
|
113
|
+
} catch {
|
|
114
|
+
return ''
|
|
115
|
+
}
|
|
116
|
+
}
|
|
117
|
+
|
|
118
|
+
/**
|
|
119
|
+
* Run the secret scan in-process and return the same {@link HookResult} shape
|
|
120
|
+
* the shell runner produced. Exit 0 → allow; 1 → advisory (marker present);
|
|
121
|
+
* 2 → block. An internal error returns exit 1 WITHOUT the marker, which the
|
|
122
|
+
* classifier maps to fail-closed.
|
|
123
|
+
*/
|
|
124
|
+
export function runSecretScanInProcess(payload: HookPayload): HookResult {
|
|
125
|
+
try {
|
|
126
|
+
const input = scannerInput(payload)
|
|
127
|
+
if (input === '') return { code: 0, stdout: '', stderr: '', signal: null, timedOut: false }
|
|
128
|
+
|
|
129
|
+
const stderr: string[] = []
|
|
130
|
+
let high: string | null = null
|
|
131
|
+
for (const pattern of [...HIGH_CONFIDENCE_RES, BIFROST_TOKEN_RE]) {
|
|
132
|
+
const match = firstMatch(input, pattern)
|
|
133
|
+
if (match !== null) {
|
|
134
|
+
stderr.push(
|
|
135
|
+
`${SECRET_SCAN_MARKER} High-confidence secret detected: ${maskSecretMatch(match)} (pattern: ${pattern.source})`,
|
|
136
|
+
)
|
|
137
|
+
high ??= match
|
|
138
|
+
}
|
|
139
|
+
}
|
|
140
|
+
let low = false
|
|
141
|
+
for (const pattern of [...LOW_CONFIDENCE_RES, BIFROST_HEADER_RE]) {
|
|
142
|
+
const match = firstMatch(input, pattern)
|
|
143
|
+
if (match !== null) {
|
|
144
|
+
stderr.push(
|
|
145
|
+
`${SECRET_SCAN_MARKER} Potential secret detected (low confidence): ${maskSecretMatch(match)} (pattern: ${pattern.source})`,
|
|
146
|
+
)
|
|
147
|
+
low = true
|
|
148
|
+
}
|
|
149
|
+
}
|
|
150
|
+
|
|
151
|
+
if (high !== null) {
|
|
152
|
+
stderr.push(
|
|
153
|
+
`${SECRET_SCAN_MARKER} High-confidence hardcoded secret detected. Tool call will be BLOCKED.`,
|
|
154
|
+
)
|
|
155
|
+
return {
|
|
156
|
+
code: 2,
|
|
157
|
+
stdout: '',
|
|
158
|
+
stderr: `${stderr.join('\n')}\n`,
|
|
159
|
+
signal: null,
|
|
160
|
+
timedOut: false,
|
|
161
|
+
}
|
|
162
|
+
}
|
|
163
|
+
if (low) {
|
|
164
|
+
stderr.push(
|
|
165
|
+
`${SECRET_SCAN_MARKER} Possible secret detected. Remove it and use environment variables or a vault.`,
|
|
166
|
+
)
|
|
167
|
+
return {
|
|
168
|
+
code: 1,
|
|
169
|
+
stdout: '',
|
|
170
|
+
stderr: `${stderr.join('\n')}\n`,
|
|
171
|
+
signal: null,
|
|
172
|
+
timedOut: false,
|
|
173
|
+
}
|
|
174
|
+
}
|
|
175
|
+
return { code: 0, stdout: '', stderr: '', signal: null, timedOut: false }
|
|
176
|
+
} catch {
|
|
177
|
+
// Fail-closed: a verdict we cannot trust denies. Exit 1 with no marker.
|
|
178
|
+
return {
|
|
179
|
+
code: 1,
|
|
180
|
+
stdout: '',
|
|
181
|
+
stderr: 'in-process secret scan failed',
|
|
182
|
+
signal: null,
|
|
183
|
+
timedOut: false,
|
|
184
|
+
}
|
|
185
|
+
}
|
|
186
|
+
}
|
|
187
|
+
|
|
188
|
+
// ─── Delegation audit log ───────────────────────────────────────────────
|
|
189
|
+
|
|
190
|
+
/** Same resolution order as the shell scripts' LOG_DIR block. */
|
|
191
|
+
export function resolveDelegationLogDir(
|
|
192
|
+
env: Record<string, string | undefined> = process.env,
|
|
193
|
+
): string {
|
|
194
|
+
if (env.LOG_DIR) return env.LOG_DIR
|
|
195
|
+
if (env.XDG_STATE_HOME) return join(env.XDG_STATE_HOME, 'pantheon', 'hooks')
|
|
196
|
+
if (env.PANTHEON_HOME) return join(env.PANTHEON_HOME, 'logs', 'agent-sessions')
|
|
197
|
+
if (env.PANTHEON_PROJECT) return join(env.PANTHEON_PROJECT, 'logs', 'agent-sessions')
|
|
198
|
+
return join('logs', 'agent-sessions')
|
|
199
|
+
}
|
|
200
|
+
|
|
201
|
+
function delegationTimestamp(now: number): string {
|
|
202
|
+
return `${new Date(now).toISOString().slice(0, 19)}Z`
|
|
203
|
+
}
|
|
204
|
+
|
|
205
|
+
/** Append one JSON audit line; best-effort, never throws. */
|
|
206
|
+
export function appendDelegationAuditLine(line: Record<string, unknown>): boolean {
|
|
207
|
+
try {
|
|
208
|
+
const dir = resolveDelegationLogDir()
|
|
209
|
+
mkdirSync(dir, { recursive: true })
|
|
210
|
+
appendFileSync(join(dir, 'delegations.log'), `${JSON.stringify(line)}\n`, 'utf8')
|
|
211
|
+
return true
|
|
212
|
+
} catch {
|
|
213
|
+
return false
|
|
214
|
+
}
|
|
215
|
+
}
|
|
216
|
+
|
|
217
|
+
/** Build the `SubagentStart` audit line (same fields as the shell script). */
|
|
218
|
+
export function delegationStartLine(fields: {
|
|
219
|
+
agent: string
|
|
220
|
+
task: string
|
|
221
|
+
sessionID: string
|
|
222
|
+
delegationID: string
|
|
223
|
+
now: number
|
|
224
|
+
}): Record<string, unknown> {
|
|
225
|
+
return {
|
|
226
|
+
event: 'SubagentStart',
|
|
227
|
+
timestamp: delegationTimestamp(fields.now),
|
|
228
|
+
agent: fields.agent,
|
|
229
|
+
task: fields.task === '' ? fields.agent : fields.task,
|
|
230
|
+
session_id: fields.sessionID,
|
|
231
|
+
delegation_id: fields.delegationID,
|
|
232
|
+
}
|
|
233
|
+
}
|
|
234
|
+
|
|
235
|
+
/** Build the `SubagentStop` audit line (same fields, duration as a number). */
|
|
236
|
+
export function delegationStopLine(fields: {
|
|
237
|
+
agent: string
|
|
238
|
+
status: string
|
|
239
|
+
reason: string
|
|
240
|
+
task: string
|
|
241
|
+
sessionID: string
|
|
242
|
+
result: unknown
|
|
243
|
+
delegationID: string
|
|
244
|
+
taskID: string
|
|
245
|
+
durationMs: number | null
|
|
246
|
+
now: number
|
|
247
|
+
}): Record<string, unknown> {
|
|
248
|
+
return {
|
|
249
|
+
event: 'SubagentStop',
|
|
250
|
+
timestamp: delegationTimestamp(fields.now),
|
|
251
|
+
agent: fields.agent,
|
|
252
|
+
status: fields.status,
|
|
253
|
+
reason: fields.reason,
|
|
254
|
+
task: fields.task === '' ? fields.agent : fields.task,
|
|
255
|
+
session_id: fields.sessionID,
|
|
256
|
+
result: fields.result ?? null,
|
|
257
|
+
delegation_id: fields.delegationID,
|
|
258
|
+
...(fields.taskID === '' ? {} : { task_id: fields.taskID }),
|
|
259
|
+
...(fields.durationMs === null ? {} : { duration_ms: fields.durationMs }),
|
|
260
|
+
}
|
|
261
|
+
}
|
|
262
|
+
|
|
263
|
+
// ─── Formatter applicability gate ───────────────────────────────────────
|
|
264
|
+
|
|
265
|
+
/** Extensions `format-multi-language.sh` actually formats; others no-op. */
|
|
266
|
+
const FORMATTABLE_EXTENSIONS = new Set([
|
|
267
|
+
'.py',
|
|
268
|
+
'.ts',
|
|
269
|
+
'.tsx',
|
|
270
|
+
'.js',
|
|
271
|
+
'.jsx',
|
|
272
|
+
'.mjs',
|
|
273
|
+
'.cjs',
|
|
274
|
+
'.json',
|
|
275
|
+
'.md',
|
|
276
|
+
'.yml',
|
|
277
|
+
'.yaml',
|
|
278
|
+
'.toml',
|
|
279
|
+
])
|
|
280
|
+
|
|
281
|
+
function formattablePath(value: unknown): boolean {
|
|
282
|
+
if (typeof value !== 'string' || value.length === 0) return false
|
|
283
|
+
const dot = value.lastIndexOf('.')
|
|
284
|
+
if (dot <= value.lastIndexOf('/')) return false
|
|
285
|
+
return FORMATTABLE_EXTENSIONS.has(value.slice(dot).toLowerCase())
|
|
286
|
+
}
|
|
287
|
+
|
|
288
|
+
/**
|
|
289
|
+
* Whether `format-multi-language.sh` would do anything for this tool call.
|
|
290
|
+
* The script no-ops for every other tool; skipping the spawn for those is the
|
|
291
|
+
* whole point of this gate.
|
|
292
|
+
*/
|
|
293
|
+
export function formatterApplies(tool: string, input: unknown): boolean {
|
|
294
|
+
const normalized = tool.trim().toLowerCase()
|
|
295
|
+
if (normalized === 'edit' || normalized === 'write' || normalized === 'hashline_edit') {
|
|
296
|
+
if (input !== null && typeof input === 'object') {
|
|
297
|
+
return formattablePath((input as Record<string, unknown>).file_path)
|
|
298
|
+
}
|
|
299
|
+
return false
|
|
300
|
+
}
|
|
301
|
+
if (normalized === 'bash') {
|
|
302
|
+
// The script scans the command for rewritten paths; the conservative gate
|
|
303
|
+
// lets bash through unchanged (rare, and correctness beats micro-gains).
|
|
304
|
+
return true
|
|
305
|
+
}
|
|
306
|
+
return false
|
|
307
|
+
}
|