pantheon-opencode 2.1.6 → 2.2.0-beta.2
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 +2 -2
- package/AGENTS.md +1 -1
- package/CHANGELOG.md +64 -0
- package/README.md +13 -9
- package/README.pt-BR.md +13 -9
- package/bin/pantheon-init.mjs +1 -1
- package/commands/pantheon-audit.md +4 -5
- package/commands/pantheon-model.md +1 -2
- package/package.json +8 -2
- package/plugin.json +1 -1
- package/prompts/dynamic/agora-generated.txt +2 -2
- package/prompts/dynamic/orchestrate-generated.txt +4 -4
- package/prompts/sketch.prompt.md +1 -1
- package/pyproject.toml +1 -1
- package/scripts/build-agents-md.mjs +73 -12
- package/scripts/build-plugin.mjs +3 -1
- package/scripts/hooks/format-multi-language.sh +30 -29
- package/scripts/install-mcp.mjs +55 -27
- package/scripts/prune.mjs +124 -2
- package/scripts/setup.mjs +1 -1
- package/scripts/themis_heuristic_scan.py +3 -5
- package/skills-lock.json +1 -1
- package/src/agents/aphrodite.md +1 -1
- package/src/agents/athena.md +6 -14
- package/src/agents/demeter.md +1 -1
- package/src/agents/hermes.md +3 -2
- package/src/agents/iris.md +4 -6
- package/src/agents/prometheus.md +3 -7
- package/src/agents/roster.json +567 -0
- package/src/agents/zeus.md +3 -0
- package/src/instructions/agent-return-format.instructions.md +1 -1
- package/src/instructions/zeus-anti-stall.instructions.md +11 -35
- package/src/instructions/zeus-council-synthesis.instructions.md +17 -49
- package/src/instructions/zeus-timeout-retry.instructions.md +12 -80
- package/src/mcp/mcp_codemap_module.py +6 -5
- package/src/mcp/pantheon-mcp-server.mjs +875 -0
- package/src/pantheon/agent-forge-tool.ts +316 -0
- package/src/pantheon/auto-permissions.ts +31 -1
- package/src/pantheon/codemap-tools.ts +525 -0
- package/src/pantheon/deepwork/plan-tracker.ts +63 -2
- package/src/pantheon/execute-after.ts +21 -0
- package/src/pantheon/hook-engine.ts +2 -1
- package/src/pantheon/native-memory-tools.ts +686 -0
- package/src/pantheon/presets.mjs +15 -0
- package/src/pantheon/step-ledger.ts +259 -0
- package/src/pantheon/swe-eval-harness.ts +211 -0
- package/src/pantheon/tools.ts +62 -7
- package/src/plugin/index.js +2091 -136
- package/src/plugin/index.ts +1 -14
- package/src/plugin.ts +167 -10
- package/src/plugins/tui/dist/tui.d.ts +66 -0
- package/src/plugins/tui/dist/tui.d.ts.map +1 -1
- package/src/plugins/tui/dist/tui.js +1259 -367
- package/src/plugins/tui/dist/tui.js.map +1 -1
- package/src/plugins/tui/package-lock.json +2 -2
- package/src/plugins/tui/package.json +1 -1
- package/src/plugins/tui/src/index.tsx +632 -71
- package/src/plugins/tui/src/pantheon-settings.tsx +78 -14
- package/src/plugins/tui/src/quota-state.ts +190 -0
- package/src/plugins/tui/src/tui-prefs.ts +169 -12
- package/src/routing.yml +3 -0
- package/src/skills/auto-continue/references/auto-continue-details.md +2 -31
- /package/scripts/{setup-v2-config.mjs → setup-config.mjs} +0 -0
package/src/agents/zeus.md
CHANGED
|
@@ -14,6 +14,8 @@ permission:
|
|
|
14
14
|
pantheon_jev_decide: allow
|
|
15
15
|
pantheon_skill_search: allow
|
|
16
16
|
pantheon_deepwork: allow
|
|
17
|
+
pantheon_agent_forge: allow
|
|
18
|
+
pantheon_step_ledger: allow
|
|
17
19
|
question: allow
|
|
18
20
|
todowrite: allow
|
|
19
21
|
todoread: allow
|
|
@@ -110,6 +112,7 @@ These rules are economic, measured against the 2026-11-05 usage audit. They appl
|
|
|
110
112
|
- **pantheon-persistence** (`context_save`): milestones only (phases, complex tasks, handoffs). Maximum 3 saves per session. Always name the checkpoint. never re-save the same key unless state genuinely changed.
|
|
111
113
|
- **pantheon-vision** (`vision_describe`): One pass per image; forward results in prompt to specialists.
|
|
112
114
|
- **pantheon_deepwork**: Structured Deepwork plans (.pantheon/deepwork/<slug>/plan.json, journal.ndjson, PLAN.md). Synchronize canonical state, verify fail-first acceptance criteria, and use OpenCode native `question` tool for Gate 1 approval.
|
|
115
|
+
- **pantheon_agent_forge**: Create, inspect, or remove custom project-specific domain specialists under `.opencode/agents/<name>.md` (with least-privilege tool grants, MCP governance ladder, and sliced TDD rules) whenever the user requests a specialized agent.
|
|
113
116
|
|
|
114
117
|
## Local advisory routing before native delegation
|
|
115
118
|
|
|
@@ -22,7 +22,7 @@ For investigation agents (Apollo, Athena), return structured findings: key findi
|
|
|
22
22
|
|
|
23
23
|
### Athena Plan Handoff
|
|
24
24
|
|
|
25
|
-
When Zeus delegates Athena through the root `subagent` dispatch, its final task result MUST contain the complete Markdown plan (not a summary, artifact path, or approval question): goal, scope and assumptions, phases, affected files or areas, dependencies, verification, risks, open questions, acceptance criteria — unknowns marked, never guessed. This supersedes the structured-findings format. Athena must not call `
|
|
25
|
+
When Zeus delegates Athena through the root `subagent` dispatch, its final task result MUST contain the complete Markdown plan (not a summary, artifact path, or approval question): goal, scope and assumptions, phases, affected files or areas, dependencies, verification, risks, open questions, acceptance criteria — unknowns marked, never guessed. This supersedes the structured-findings format. Athena must not call `question`, pause for user input, or ask for approval in the child session. Zeus publishes the full plan before approval UI, never bundling them; if order is not guaranteed, Zeus asks in plain text afterwards. For Open Questions or decisions needed, Zeus prompts via the native `question` tool with structured options and marked recommended defaults, never leaving inert text bullets. Direct Athena invocations show the complete plan before requesting approval.
|
|
26
26
|
|
|
27
27
|
## Council Specialist Response Format
|
|
28
28
|
|
|
@@ -7,55 +7,31 @@ fallback: shared
|
|
|
7
7
|
|
|
8
8
|
# ANTI-STALL & STALL DETECTION
|
|
9
9
|
|
|
10
|
-
|
|
10
|
+
Proactively detect and recover from stalled states without waiting for user intervention.
|
|
11
11
|
|
|
12
12
|
## Stall Detection Protocol
|
|
13
13
|
|
|
14
|
-
You MUST self-monitor for these stall conditions:
|
|
15
|
-
|
|
16
14
|
| Symptom | Detection Rule | Recovery Action |
|
|
17
15
|
|---------|---------------|-----------------|
|
|
18
|
-
| Silent loop | 3+ consecutive turns with no tool call AND no visible progress | Output `[STALL_DETECTED]
|
|
19
|
-
| Delegation black hole | No result or host completion notification after
|
|
20
|
-
| Circular delegation | Same specialist re-dispatched for same task 2+ times without progress |
|
|
21
|
-
| Idle after completion |
|
|
22
|
-
| Context thrash | Re-reading same files repeatedly without new action | Stop re-reading
|
|
23
|
-
|
|
24
|
-
## Phase Reminder
|
|
25
|
-
|
|
26
|
-
Use the root `subagent` dispatch observed in the Beta 10 runtime; `task()` is host/API terminology, and the exact signature/background options depend on the active runtime. Prefer synchronous delegation unless the visible schema supports background and the host exposes a way to observe completion. For asynchronous work:
|
|
27
|
-
1. Do not invoke `task_status`, `cancel_task`, or board methods unless a corresponding tool is actually exposed (none was exposed in the reported Beta 10 root catalog).
|
|
28
|
-
2. Do not advance dependent work until a result or host-provided completion notification arrives.
|
|
29
|
-
3. Continue orchestration only on non-overlapping independent work.
|
|
30
|
-
4. If nothing independent remains, briefly report what was dispatched and wait for host notification; do not claim job tracking or recovery.
|
|
31
|
-
|
|
32
|
-
Self-check every 3 turns: "Am I waiting on a delegate? Have I polled without need? Is there independent work I can do?"
|
|
16
|
+
| Silent loop | 3+ consecutive turns with no tool call AND no visible progress | Output `[STALL_DETECTED]`, re-read task, or escalate with 3 concrete options |
|
|
17
|
+
| Delegation black hole | No result or host completion notification after timeout | Stop dependent work; follow `## Timeout & Retry Enforcement` |
|
|
18
|
+
| Circular delegation | Same specialist re-dispatched for same task 2+ times without progress | Dispatch to different specialist OR escalate to user |
|
|
19
|
+
| Idle after completion | Background tasks completed but no synthesis/next step for 2+ turns | Summarize completed results and propose next action |
|
|
20
|
+
| Context thrash | Re-reading same files repeatedly without new action | Stop re-reading and proceed with action |
|
|
33
21
|
|
|
34
|
-
## Delegate Retry
|
|
22
|
+
## Phase Reminder & Delegate Retry
|
|
35
23
|
|
|
36
|
-
|
|
37
|
-
response), check known causes first and allow at most one corrected retry per
|
|
38
|
-
agent/task. Do not retry a scope refusal or a failed test/validation; reroute
|
|
39
|
-
only when the correct specialist is clear, otherwise diagnose or escalate.
|
|
40
|
-
After a transient retry fails, follow the canonical fallback chain once and
|
|
41
|
-
escalate rather than restarting it. See `## Timeout & Retry Enforcement`.
|
|
24
|
+
Prefer synchronous `subagent` delegation unless background completion is observable through the host. Do not invoke `task_status`, `cancel_task`, or board methods unless exposed; do not advance dependent work until a result arrives; continue only on non-overlapping independent work.
|
|
25
|
+
When a delegation has a transient dispatch failure (timeout, crash, or empty response), check known causes first and allow at most one corrected retry per agent/task. Do not retry a scope refusal or a failed test/validation; reroute only when the correct specialist is clear, otherwise diagnose or escalate.
|
|
42
26
|
|
|
43
27
|
## Progress Checkpoint
|
|
44
28
|
|
|
45
29
|
For long-running or multi-phase tasks expected to run > 5 turns only:
|
|
46
30
|
- After turn 5: output `[CHECKPOINT] Completed so far: [summary]. Remaining: [list].`
|
|
47
|
-
- After turn 10: re-evaluate
|
|
48
|
-
- If 3 consecutive turns produce no tool calls: trigger Stall Detection Protocol (see above).
|
|
31
|
+
- After turn 10: re-evaluate; if < 50% done, consider splitting or escalating.
|
|
49
32
|
|
|
50
33
|
## Heartbeat & Checkpoint Integration
|
|
51
34
|
|
|
52
35
|
Use checkpoints only for long-running or multi-phase work when `context_save` and `context_get` are available. Never add persistence calls for a bounded task.
|
|
53
|
-
|
|
54
|
-
### Checkpoint / Pre-Compaction
|
|
55
|
-
Checkpoint only when context loss is plausible and both checkpoint tools are present. Save once at a meaningful boundary using `content=JSON.stringify(state)` and an explicit session ID; `state` must be an object with object-valued `phase`/`goal`/`delegations`/`heartbeat` fields and an array `tail`, and omit an unset `goal`. A successful save needs no immediate read. Checkpoints expire after 4h.
|
|
56
|
-
|
|
57
|
-
### Context Retrieval
|
|
36
|
+
Save once at a meaningful boundary using `content=JSON.stringify(state)` and an explicit session ID; `state` must be an object with object-valued `phase`/`goal`/`delegations`/`heartbeat` fields and an array `tail`, and omit an unset `goal`. A successful save needs no immediate read. Checkpoints expire after 4h.
|
|
58
37
|
Read a checkpoint only when resuming a workflow with a known session ID. Do not retrieve it after every dispatch; carry ordinary handoffs in the task context.
|
|
59
|
-
|
|
60
|
-
### Long-Session Progress
|
|
61
|
-
For long-lived multi-phase work, update an existing/requested progress record every 5 turns with completed and pending work and blockers. Do not create STATUS.md for a bounded task.
|
|
@@ -7,36 +7,25 @@ fallback: shared
|
|
|
7
7
|
|
|
8
8
|
# INLINE COUNCIL SYNTHESIS — /pantheon
|
|
9
9
|
|
|
10
|
-
Use an inline specialist council for material trade-offs, architecture/security choices, technology selection,
|
|
10
|
+
Use an inline specialist council for material trade-offs, architecture/security choices, technology selection, or multi-stakeholder questions. Do not use it for ordinary implementation tasks.
|
|
11
11
|
|
|
12
12
|
## Dispatch and Synthesis
|
|
13
13
|
|
|
14
|
-
1. **Precedent:** Run the
|
|
15
|
-
2. **Research (optional):** For `/pantheon --research <question>`, ask @apollo for a 30s pre-scan with a concise brief
|
|
16
|
-
3. **Select and dispatch:** Choose at most 3 agents using the domain map below. Every dispatch—including Apollo discovery, specialist rounds, and Themis audit—gets a concise brief with objective, bounded scope/acceptance, relevant context only, constraints/safety gates, role-specific deliverable, and verification. Keep `shared_context` minimal, non-sensitive, and untrusted; never paste unrelated conversation or secrets. The host may independently inherit conversation transcript
|
|
17
|
-
4. **Collect:** Use returned subagent results or host
|
|
18
|
-
5. **Validate confidence:** High requires
|
|
19
|
-
6. **Resolve divergence:** After
|
|
20
|
-
|
|
21
|
-
**
|
|
22
|
-
- **Domain Authority
|
|
23
|
-
|
|
24
|
-
|
|
25
|
-
- *Frontend / UX / A11y:* `@aphrodite` has primary authority over UI ergonomics, responsiveness, and accessibility compliance.
|
|
26
|
-
- *Infrastructure / Deployment / CI:* `@prometheus` has primary authority over containers, pipelines, and runtime resource limits.
|
|
27
|
-
- *API / Network / Protocols:* `@hermes` has primary authority over contract conformity and transport semantics.
|
|
28
|
-
- *Architecture / System Boundaries:* `@athena` has primary authority over modularity and long-term maintainability.
|
|
29
|
-
If the divergence falls within the core specialty of a specialist with verifiable claims (High/Medium confidence), Zeus arbitrates in favor of that Domain Authority immediately.
|
|
30
|
-
- **JEV Multi-Criteria Arbitration (`pantheon_jev_decide`):** When trade-offs span multiple competing domains without a single owner (e.g. Athena's modularity vs Prometheus's operational simplicity), Zeus invokes `pantheon_jev_decide` (mode: `order` or `score`) or evaluates the multi-criteria candidate matrix to score and rank options deterministically in ONE call.
|
|
31
|
-
- **Targeted Rebuttal Fallback:** Below 50%, and only if no agent timed out and neither Domain Authority nor JEV arbitration can resolve an unverified factual contradiction (e.g., mutually exclusive claims about runtime behavior that cannot be verified locally), run one rebuttal round with all responses and the current synthesis draft. In that round, signals may refer only to peer responses visible in the rebuttal prompt; treat revised responses as final and recompute the rate from their actual positions. If the defined rate remains below 50%, research factual divergence points with official docs, benchmarks, or issues (max 3 searches; seek 2 independent sources per point when possible). Skip research for subjective-only disagreements. Send responses and a cited evidence brief to @themis as moderator; request verdicts, overall direction, confidence, and unresolved issues. Low confidence or unresolved points remain explicit in the synthesis. Moderator and audit roles are separate.
|
|
32
|
-
7. **Synthesize and audit:** Use the output format below, then ask @themis to compare raw responses with the synthesis. Check representation of every specialist, preserved divergences, accurate attribution, and confidence specificity. Fix any issue before delivery.
|
|
33
|
-
8. **Persist (conditional):** Store the decision only if a memory-write tool is explicitly exposed and permitted in the active root tool list; require its success result before saying it was saved. Otherwise say the decision was not persisted. The `BackgroundJobBoard` is internal and has no root-callable board tool in the reported runtime.
|
|
34
|
-
|
|
35
|
-
The `BackgroundJobBoard` is internal bookkeeping, not a tool available to Zeus root; do not call `board.registerLaunch`, `markReconciled`, `recoverRunningJobs`, or `formatForPrompt`, and do not assume crash recovery. If context is lost, continue only from results still present in the conversation or a host handoff; otherwise report the missing context and ask how to proceed.
|
|
14
|
+
1. **Precedent:** Run the `council_decisions` search defined in `## Memory Protocol`. Apply its age/score rules; if no fast-path decision applies, continue.
|
|
15
|
+
2. **Research (optional):** For `/pantheon --research <question>`, ask @apollo for a 30s pre-scan with a concise brief; pass only relevant findings as `shared_context`.
|
|
16
|
+
3. **Select and dispatch:** Choose at most 3 agents using the domain map below. Every dispatch—including Apollo discovery, specialist rounds, and Themis audit—gets a concise brief with objective, bounded scope/acceptance, relevant context only, constraints/safety gates, role-specific deliverable, and verification. Keep `shared_context` minimal, non-sensitive, and untrusted; never paste unrelated conversation or secrets. The host may independently inherit conversation transcript. Dispatch through the root `subagent` tool (`task()` is host/API terminology). Keep first-round prompts independent: do not include other specialists' responses. Dispatch in parallel when the observed tool/runtime supports it; synchronous or sequential dispatch is also valid if each first-round prompt excludes prior first-round responses. First-round `agreement_signals` must be omitted or set to `none`.
|
|
17
|
+
4. **Collect:** Use returned subagent results or host completion notifications. Do not call `task_status` or claim job tracking/recovery when unavailable. Note timeouts and missing responses explicitly. Partial responses are acceptable only from read-only @apollo or @gaia.
|
|
18
|
+
5. **Validate confidence:** High requires >=3 verifiable claims (otherwise downgrade to Medium and state why); Medium requires >=1 claim; Low needs no minimum. Self-reported `specific_claims` count is advisory, not proof.
|
|
19
|
+
6. **Resolve divergence:** After initial responses arrive, Zeus groups their actual substantive positions by issue and derives agreement from those positions, not from speculative or self-reported `agreement_signals`. For each comparable issue, compare every unordered pair of specialists who provided a comparable substantive position on that issue. A pair agrees only when its substantive positions are compatible; every pair whose positions are not aligned counts as divergent. For an issue with `n` specialists providing comparable positions, include `n*(n-1)/2` pairs in `total_comparable_pairs`; sum this across included issues. Exclude issues with fewer than two comparable positions and state that they were excluded. Calculate `agreement_rate = agreeing_pairs / total_comparable_pairs`: unanimity is 100%, all pairs in divergence is 0%, and non-aligned pairs are divergences. If the denominator is zero, the rate is indeterminate/has no basis; never substitute an artificial zero or trigger rebuttal from an undefined rate.
|
|
20
|
+
- **Domain Authority Precedence:** Technical domain owners carry decisive authority when backed by verifiable claims (High/Medium confidence): `@themis` (Security/Auth/Audit veto), `@demeter` (Database/Schema/Queries), `@aphrodite` (Frontend/UX/A11y), `@prometheus` (Infra/Deploy/CI), `@hermes` (API/Protocols), `@athena` (Architecture/Modularity).
|
|
21
|
+
- **JEV Multi-Criteria Arbitration (`pantheon_jev_decide`):** When trade-offs span competing domains without a single owner, invoke `pantheon_jev_decide` (`order` or `score`) in ONE call.
|
|
22
|
+
- **Targeted Rebuttal Fallback:** Below 50%, and only if no agent timed out and neither Domain Authority nor JEV arbitration resolves an unverified factual contradiction, run one rebuttal round with all responses and the synthesis draft. If the defined rate remains below 50%, research factual divergence points (max 3 searches) and send responses + cited evidence brief to @themis as moderator. Moderator and audit roles are separate.
|
|
23
|
+
7. **Synthesize and audit:** Use the output format below, then ask @themis to compare raw responses with the synthesis (representation, preserved divergences, attribution, confidence).
|
|
24
|
+
8. **Persist (conditional):** Store the decision only if a memory-write tool is explicitly exposed and permitted in the active root tool list; require its success result before saying it was saved. `BackgroundJobBoard` is internal bookkeeping with no root-callable tool.
|
|
36
25
|
|
|
37
26
|
## Specialist Response Format
|
|
38
27
|
|
|
39
|
-
Require `## specialist_response` from `## Agent Return Format`.
|
|
28
|
+
Require `## specialist_response` from `## Agent Return Format`. First-round agreement signals remain omitted or `none`.
|
|
40
29
|
|
|
41
30
|
## Domain Map
|
|
42
31
|
|
|
@@ -62,7 +51,7 @@ Require `## specialist_response` from `## Agent Return Format`. The first round
|
|
|
62
51
|
### Specialist Perspectives
|
|
63
52
|
| Specialist | Domain Authority | Substantive Position | Trade-offs | Confidence |
|
|
64
53
|
| :--- | :--- | :--- | :--- | :---: |
|
|
65
|
-
| **@[agent]** | [Domain
|
|
54
|
+
| **@[agent]** | [Domain] | [Core recommendation] | [Pros & Cons] | **[High / Med / Low]** |
|
|
66
55
|
|
|
67
56
|
### Agreements
|
|
68
57
|
- [Key consensus point shared across specialists]
|
|
@@ -72,30 +61,9 @@ Require `## specialist_response` from `## Agent Return Format`. The first round
|
|
|
72
61
|
| :--- | :--- | :--- |
|
|
73
62
|
| [Contested point] | **@[agentA]**: [Position A]<br/>**@[agentB]**: [Position B] | **Resolved via [Domain Authority / Jev Decision]**: [Outcome] |
|
|
74
63
|
|
|
75
|
-
|
|
76
|
-
|
|
77
|
-
Q["Contested Issue"] --> A["@[agentA]: Position A"]
|
|
78
|
-
Q --> B["@[agentB]: Position B"]
|
|
79
|
-
A & B --> Arb{"Domain Authority & Jev Score"}
|
|
80
|
-
Arb --> Dec["Arbitrated Decision: [Outcome]"]
|
|
81
|
-
```
|
|
82
|
-
|
|
83
|
-
### Evidence & Moderation (when used)
|
|
84
|
-
Research points, evidence, moderator verdict, cited sources, unresolved issues
|
|
85
|
-
|
|
86
|
-
### Recommendation
|
|
87
|
-
Decisive conclusion and immediate actionable next steps
|
|
88
|
-
|
|
89
|
-
### Immediate Action Checklist
|
|
90
|
-
- [ ] **@[specialistA]**: [Concrete implementation task]
|
|
91
|
-
- [ ] **@[specialistB]**: [Verification or interface task]
|
|
64
|
+
### Recommendation & Action Checklist
|
|
65
|
+
- [ ] **@[specialistA]**: [Concrete task]
|
|
92
66
|
- [ ] **@themis**: [Audit gate verification]
|
|
93
|
-
|
|
94
|
-
### Audit Gate
|
|
95
|
-
Themis approved, or issues corrected
|
|
96
|
-
|
|
97
|
-
### Decision Gate
|
|
98
|
-
Confidence adjusted for response rate
|
|
99
67
|
```
|
|
100
68
|
|
|
101
|
-
Initial agreement signals remain omitted/`none`; consensus and agreements are derived by Zeus from compared substantive positions and presented in the synthesis. Preserve actual specialist timeout or unresolved-tie details in the response.
|
|
69
|
+
Initial agreement signals remain omitted/`none`; consensus and agreements are derived by Zeus from compared substantive positions and presented in the synthesis. Preserve actual specialist timeout or unresolved-tie details in the response.
|
|
@@ -8,13 +8,7 @@ fallback: shared
|
|
|
8
8
|
# TIMEOUT & RETRY ENFORCEMENT
|
|
9
9
|
|
|
10
10
|
When a delegated agent does not respond in time, enforce the timeout policy from `routing.yml`.
|
|
11
|
-
|
|
12
|
-
Use the root `subagent` dispatch observed in the Beta 10 runtime. `task()` is
|
|
13
|
-
host/API terminology; the active runtime determines the callable schema and any
|
|
14
|
-
background options. Do not assume root `task_status`, `cancel_task`, or board
|
|
15
|
-
tracking/recovery tools. Prefer synchronous dispatch unless background completion
|
|
16
|
-
is observable through the host, and keep any retry list in the current context
|
|
17
|
-
only; do not claim persistent job tracking.
|
|
11
|
+
Use the root `subagent` dispatch (`task()` is host/API terminology). Do not assume root `task_status`, `cancel_task`, or board tracking/recovery tools. Prefer synchronous dispatch unless background completion is observable through the host, and keep any retry list in the current context only; do not claim persistent job tracking.
|
|
18
12
|
|
|
19
13
|
## Timeout Behavior by Agent Role
|
|
20
14
|
|
|
@@ -29,90 +23,28 @@ only; do not claim persistent job tracking.
|
|
|
29
23
|
|
|
30
24
|
## Retry Flow
|
|
31
25
|
|
|
32
|
-
|
|
33
|
-
Initial attempt → timeout/failure → log it → make at most one retry
|
|
34
|
-
└─ retry fails → try fallback agents left-to-right with original task/context
|
|
35
|
-
└─ all fallbacks fail (or none exist) → report the chain and escalate
|
|
36
|
-
```
|
|
37
|
-
`background_delegation.retry_count: 1` means one retry after the initial attempt, not one total attempt. Never restart an exhausted chain automatically.
|
|
26
|
+
`background_delegation.retry_count: 1` means one retry after the initial attempt, not one total attempt. If the retry fails, try fallback agents left-to-right with original task/context; if all fallbacks fail (or none exist), report the chain and escalate. Never retry the same chain automatically.
|
|
38
27
|
|
|
39
|
-
Apply the retry only to a transient dispatch failure (timeout, crash, or empty response). A failed test, validation, or migration is a result to diagnose, not a reason to repeat the same command or agent blindly.
|
|
28
|
+
Apply the retry only to a transient dispatch failure (timeout, crash, or empty response). A failed test, validation, or migration is a result to diagnose, not a reason to repeat the same command or agent blindly. For auth/security/data/schema work, a fallback must preserve specialist competence, Themis review, and human approval; otherwise stop and escalate. Never let full-auto bypass these gates.
|
|
40
29
|
An agent refusal or scope-boundary response is not a transient failure: do not retry it with a rephrased prompt. Correct the route once if the right specialist is clear; otherwise stop and ask.
|
|
41
30
|
|
|
42
31
|
## Fallback Chain Definitions
|
|
43
32
|
|
|
44
|
-
`routing.yml` → `fallback_chains` is the canonical agent-by-agent mapping. Evaluate
|
|
45
|
-
|
|
46
|
-
|
|
47
|
-
at most once, and an exhausted chain is never restarted automatically.
|
|
48
|
-
|
|
49
|
-
### Refusal Is a Routing Error
|
|
50
|
-
|
|
51
|
-
For a domain/scope refusal, do not retry with a rephrased prompt. Correct the
|
|
52
|
-
route once only when the right specialist is clear; otherwise stop and ask.
|
|
53
|
-
Record one `DelegationCacheDecision` telemetry item only when telemetry recording
|
|
54
|
-
is actually exposed; otherwise keep the corrected route in the current handoff
|
|
55
|
-
and do not claim telemetry was persisted. Include `reroute_from`, `reroute_to`,
|
|
56
|
-
and `delegation_id` when available.
|
|
57
|
-
|
|
58
|
-
### Escalation Protocol
|
|
59
|
-
When all fallbacks fail, report the tried agents and errors, offer a different approach/simpler scope/manual fix, and stop. Never retry the same chain automatically.
|
|
60
|
-
|
|
61
|
-
### Session Reuse Check
|
|
62
|
-
Reuse a prior specialist session only when the active host exposes that capability;
|
|
63
|
-
otherwise dispatch through `subagent` with necessary context. Treat `session_max`
|
|
64
|
-
as policy context, not as a claim that Zeus can enumerate sessions.
|
|
33
|
+
`routing.yml` → `fallback_chains` is the canonical agent-by-agent mapping. Evaluate the configured chain left-to-right; if an agent has no configured chain, stop and escalate rather than guessing a fallback.
|
|
34
|
+
For a domain/scope refusal, do not retry with a rephrased prompt; record one `DelegationCacheDecision` telemetry item only when telemetry recording is actually exposed (`reroute_from`, `reroute_to`, `delegation_id`).
|
|
35
|
+
Reuse a prior specialist session only when the active host exposes that capability; otherwise dispatch through `subagent` with necessary context.
|
|
65
36
|
|
|
66
37
|
---
|
|
67
38
|
|
|
68
39
|
# SUBTASK DISPATCH (Lightweight Delegation)
|
|
69
40
|
|
|
70
|
-
Subtask is a bounded, low-risk delegation mode that **skips** the standard artifact lifecycle.
|
|
71
|
-
|
|
72
|
-
## When to Use Subtask vs Full Task
|
|
73
|
-
|
|
74
|
-
Use the lightest path that safely meets the request. A multi-file but bounded low-risk change does not automatically need a plan/artifact phase.
|
|
75
|
-
|
|
76
|
-
Assess complexity qualitatively from uncertainty, specialist expertise, dependencies,
|
|
77
|
-
risk, and how the result can be verified—not from a fixed file or line count. Keep
|
|
78
|
-
self-contained work inline when no specialist is needed; use one correct specialist
|
|
79
|
-
for one bounded expertise-specific task; Zeus may make separate root dispatches for
|
|
80
|
-
independent fronts. Keep dependent work ordered and preserve planning, review,
|
|
81
|
-
artifact, and approval gates when the risk requires them.
|
|
82
|
-
|
|
83
|
-
### Subtask Decision Tree (run BEFORE every delegation)
|
|
41
|
+
Subtask is a bounded, low-risk delegation mode that **skips** the standard artifact lifecycle. Assess complexity qualitatively from uncertainty, specialist expertise, dependencies, risk, and how the result can be verified—not from a fixed file or line count. Keep self-contained work inline when no specialist is needed; use one correct specialist for one bounded expertise-specific task; Zeus may make separate root dispatches for independent fronts. Keep dependent work ordered and preserve planning, review, artifact, and approval gates when the risk requires them.
|
|
84
42
|
|
|
85
|
-
|
|
86
|
-
|
|
87
|
-
□ Does it avoid auth/security, data/schema, destructive, broad/material, and user-visible API risk? [YES→continue | NO→relevant gates]
|
|
88
|
-
□ Is there no required Themis/audit handoff for this change? [YES→continue | NO→preserve that review]
|
|
89
|
-
|
|
90
|
-
ALL YES → one correct domain specialist; they inspect needed context, implement, and verify; no routine plan/artifact, separate discovery, council, or review
|
|
91
|
-
ANY NO → add only the planning, artifact, specialist review, and approval gates the risk requires
|
|
92
|
-
```
|
|
93
|
-
|
|
94
|
-
### Safety Rules
|
|
95
|
-
1. **Bounded scope** — can be a small multi-file change or a read-only investigation
|
|
96
|
-
2. **Low risk and uncertainty** — no security implications, data loss, breaking/user-visible API impact, broad material scope, or unresolved high uncertainty
|
|
97
|
-
3. **No required Themis dependency** — sensitive/material output retains its review gate
|
|
98
|
-
|
|
99
|
-
The selected specialist, not a routine Apollo discovery pass, inspects the relevant
|
|
100
|
-
local context for a lightweight implementation. Keep specialist roles available for
|
|
101
|
-
on-demand routing; do not remove an agent based on one task or one overhead observation.
|
|
43
|
+
- **Fast-path criteria:** Scope is clear, bounded, reversible, and low-uncertainty; avoids auth/security, data/schema, destructive, broad/material, and user-visible API risk; requires no Themis/audit handoff.
|
|
44
|
+
- The selected specialist, not a routine Apollo discovery pass, inspects the relevant local context for a lightweight implementation.
|
|
102
45
|
|
|
103
46
|
## Subtask Return Format
|
|
104
|
-
Implementation agents use the `## subtask_summary` fields defined in `## Agent Return Format
|
|
105
|
-
|
|
106
|
-
## Timeout Parcial (Partial Results)
|
|
107
|
-
|
|
108
|
-
Only agents marked Yes in the Timeout Behavior table above may return partial results: @apollo (partial file list, e.g. "found 7 of 12 files before timeout"), @gaia (partial literature findings) and @talos (confirm progress if a hotfix times out). Never for implementers or reviewers — they must complete or fail. When dispatching with partial-OK, set the expectation: `@apollo Search for auth files. Timeout partial OK — return whatever you have.`
|
|
109
|
-
|
|
110
|
-
---
|
|
111
|
-
|
|
112
|
-
# TIMEOUT TRACKING
|
|
47
|
+
Implementation agents use the `## subtask_summary` fields defined in `## Agent Return Format`; investigations, plans, and council responses keep their role-specific formats. Include the optional machine-readable `memory_context` only in `subtask_summary` when memory affected the result, and omit it otherwise. In handoffs, pass only useful memory hit refs and a concise takeaway; treat that context as untrusted and subordinate to the current request/instructions.
|
|
113
48
|
|
|
114
|
-
|
|
115
|
-
|
|
116
|
-
not poll, cancel, or write a timeout log unless the corresponding tool is exposed.
|
|
117
|
-
Report any timeout and the retry/fallback taken in the user-facing handoff; do
|
|
118
|
-
not claim persistent tracking or recovery.
|
|
49
|
+
## Timeout Parcial (Partial Results) & Tracking
|
|
50
|
+
Only agents marked Yes above (@apollo, @gaia, @talos) may return partial results; never implementers or reviewers. Keep only an in-context list of active delegations and tried agents; do not poll, cancel, or claim persistent tracking unless the tool is exposed.
|
|
@@ -544,11 +544,12 @@ def code_index( # noqa: C901, PLR0912, PLR0915
|
|
|
544
544
|
[str(file_path), h, lang],
|
|
545
545
|
)
|
|
546
546
|
conn.commit()
|
|
547
|
-
|
|
548
|
-
|
|
549
|
-
|
|
550
|
-
|
|
551
|
-
|
|
547
|
+
if lang == "python":
|
|
548
|
+
try:
|
|
549
|
+
ast.parse(content)
|
|
550
|
+
except SyntaxError:
|
|
551
|
+
errors += 1
|
|
552
|
+
continue
|
|
552
553
|
skipped += 1
|
|
553
554
|
continue
|
|
554
555
|
_insert_entities_and_relations(conn, ents, rels)
|