[integration][auto-dig] actor failure in run 33094075463 #277

New issue

Closed

opened 2026-08-27 16:37:52 +00:00 by lost-rob0t · 2 comments

lost-rob0t commented

2026-08-27 16:37:52 +00:00

(Migrated from github.com)

Auto-Dig integration failure diagnostics

Consumer: lost-rob0t/starintel-gpt-auto-dig
Workflow run: https://github.com/lost-rob0t/starintel-gpt-auto-dig/actions/runs/33094075463
Consumer head: aacbe53f9385a2c93c4192e0373cb5524077be68
Prolog-RLM ref: 7f0050f524db6de5293163241176ac4660f04349
Auto-Dig target: lost-rob0t/starintel-gpt-auto-dig#2284

Sanitized failure diagnostics

Summary: phase=planner kind=plan_parse_failed message=planner responses did not contain a valid typed plan model_calls=3 prompt_tokens=8121 completion_tokens=4307 total_tokens=12428 cost_usd=0.00595436

Structured Prolog-RLM error

{
  "attempts": 3,
  "cause": {
    "detail": {
      "$term": "missing_field",
      "args": [
        "name"
      ]
    },
    "kind": "invalid_plan",
    "message": "plan could not be normalized",
    "phase": "normalize"
  },
  "kind": "plan_parse_failed",
  "message": "planner responses did not contain a valid typed plan",
  "phase": "planner",
  "usage": {
    "completion_tokens": 4307,
    "cost_known": true,
    "cost_usd": 0.00595436,
    "model_calls": 3,
    "prompt_tokens": 8121,
    "tokens_known": true,
    "total_tokens": 12428
  }
}

Sanitized traceback / stderr tail (last 160 lines)

[auto_dig_rlm] 2026-08-27T16:36:47Z phase=start model=openai/gpt-5.6-luna reasoning_effort=max
[auto_dig_rlm] 2026-08-27T16:36:47Z phase=context_loaded chars=6431 file=agent-runs/auto-dig-prolog/2026-08-27-33094075463-1/request.md
[auto_dig_rlm] 2026-08-27T16:36:47Z phase=mcp_session_open servers=[brave,fetch]
[auto_dig_mcp] 2026-08-27T16:36:47Z session_open servers=[brave,fetch]
[auto_dig_mcp] 2026-08-27T16:36:47Z phase=registry_create state=start
[auto_dig_mcp] 2026-08-27T16:36:47Z phase=registry_create state=ok
[auto_dig_mcp] 2026-08-27T16:36:47Z phase=authority_setup state=start
[auto_dig_mcp] 2026-08-27T16:36:47Z phase=authority_setup state=ok
[auto_dig_mcp] 2026-08-27T16:36:47Z phase=mcp_run(brave) state=start
[auto_dig_mcp] 2026-08-27T16:36:47Z phase=mcp_run(brave) state=ok
[auto_dig_mcp] 2026-08-27T16:36:47Z phase=mcp_connect(brave) state=start
[auto_dig_mcp] 2026-08-27T16:36:55Z phase=mcp_connect(brave) state=ok
[auto_dig_mcp] 2026-08-27T16:36:55Z phase=mcp_import(brave) state=start
[auto_dig_mcp] 2026-08-27T16:36:55Z phase=mcp_import(brave) state=ok
[auto_dig_mcp] 2026-08-27T16:36:55Z mcp_import server=brave imported_tools=8
[auto_dig_mcp] 2026-08-27T16:36:55Z phase=mcp_run(fetch) state=start
[auto_dig_mcp] 2026-08-27T16:36:55Z phase=mcp_run(fetch) state=ok
[auto_dig_mcp] 2026-08-27T16:36:55Z phase=mcp_connect(fetch) state=start
[auto_dig_mcp] 2026-08-27T16:37:01Z phase=mcp_connect(fetch) state=ok
[auto_dig_mcp] 2026-08-27T16:37:01Z phase=mcp_import(fetch) state=start
[auto_dig_mcp] 2026-08-27T16:37:01Z phase=mcp_import(fetch) state=ok
[auto_dig_mcp] 2026-08-27T16:37:01Z mcp_import server=fetch imported_tools=6
[auto_dig_mcp] 2026-08-27T16:37:01Z phase=capability_projection state=start
[auto_dig_mcp] 2026-08-27T16:37:01Z phase=capability_projection state=ok
[auto_dig_mcp] 2026-08-27T16:37:01Z capability_projection count=8
[auto_dig_mcp] 2026-08-27T16:37:01Z session_open state=ok
[auto_dig_rlm] 2026-08-27T16:37:01Z phase=mcp_ready capability_count=8 capabilities=[tool('mcp.brave.brave_news_search'),tool('mcp.brave.brave_video_search'),tool('mcp.brave.brave_web_search'),tool('mcp.fetch.fetch_json'),tool('mcp.fetch.fetch_markdown'),tool('mcp.fetch.fetch_readable'),tool('mcp.fetch.fetch_txt'),tool('mcp.fetch.fetch_youtube_transcript')]
[auto_dig_rlm] 2026-08-27T16:37:01Z phase=rlm_start token_budget=50000 model_calls=6 tool_calls=8 recursion_depth=2 time_limit=90.0
[auto_dig_rlm] 2026-08-27T16:37:49Z phase=rlm_return
::error title=Auto-Dig Prolog-RLM failure::status=error phase=planner kind=plan_parse_failed message=planner responses did not contain a valid typed plan used=unknown limit=unknown model_calls=3 prompt_tokens=8121 completion_tokens=4307 total_tokens=12428 cost_usd=0.00595436
[auto_dig_rlm] 2026-08-27T16:37:49Z phase=outcome status=error phase=planner kind=plan_parse_failed message=planner responses did not contain a valid typed plan used=unknown limit=unknown model_calls=3 prompt_tokens=8121 completion_tokens=4307 total_tokens=12428 cost_usd=0.00595436
[auto_dig_rlm] 2026-08-27T16:37:49Z phase=result_written file=agent-runs/auto-dig-prolog/2026-08-27-33094075463-1/rlm-result.json
[auto_dig_rlm] 2026-08-27T16:37:49Z phase=trace_written file=agent-runs/auto-dig-prolog/2026-08-27-33094075463-1/rlm-trace.json
[auto_dig_rlm] 2026-08-27T16:37:49Z phase=finish exit_code=1

Diagnostics are sanitized at emission time. Raw environment dumps and raw provider payloads are never included in issue-bound output.

## Auto-Dig integration failure diagnostics - Consumer: `lost-rob0t/starintel-gpt-auto-dig` - Workflow run: https://github.com/lost-rob0t/starintel-gpt-auto-dig/actions/runs/33094075463 - Consumer head: `aacbe53f9385a2c93c4192e0373cb5524077be68` - Prolog-RLM ref: `7f0050f524db6de5293163241176ac4660f04349` - Auto-Dig target: lost-rob0t/starintel-gpt-auto-dig#2284 ## Sanitized failure diagnostics **Summary:** `phase=planner kind=plan_parse_failed message=planner responses did not contain a valid typed plan model_calls=3 prompt_tokens=8121 completion_tokens=4307 total_tokens=12428 cost_usd=0.00595436` ### Structured Prolog-RLM error ```json { "attempts": 3, "cause": { "detail": { "$term": "missing_field", "args": [ "name" ] }, "kind": "invalid_plan", "message": "plan could not be normalized", "phase": "normalize" }, "kind": "plan_parse_failed", "message": "planner responses did not contain a valid typed plan", "phase": "planner", "usage": { "completion_tokens": 4307, "cost_known": true, "cost_usd": 0.00595436, "model_calls": 3, "prompt_tokens": 8121, "tokens_known": true, "total_tokens": 12428 } } ``` ### Sanitized traceback / stderr tail (last 160 lines) ```text [auto_dig_rlm] 2026-08-27T16:36:47Z phase=start model=openai/gpt-5.6-luna reasoning_effort=max [auto_dig_rlm] 2026-08-27T16:36:47Z phase=context_loaded chars=6431 file=agent-runs/auto-dig-prolog/2026-08-27-33094075463-1/request.md [auto_dig_rlm] 2026-08-27T16:36:47Z phase=mcp_session_open servers=[brave,fetch] [auto_dig_mcp] 2026-08-27T16:36:47Z session_open servers=[brave,fetch] [auto_dig_mcp] 2026-08-27T16:36:47Z phase=registry_create state=start [auto_dig_mcp] 2026-08-27T16:36:47Z phase=registry_create state=ok [auto_dig_mcp] 2026-08-27T16:36:47Z phase=authority_setup state=start [auto_dig_mcp] 2026-08-27T16:36:47Z phase=authority_setup state=ok [auto_dig_mcp] 2026-08-27T16:36:47Z phase=mcp_run(brave) state=start [auto_dig_mcp] 2026-08-27T16:36:47Z phase=mcp_run(brave) state=ok [auto_dig_mcp] 2026-08-27T16:36:47Z phase=mcp_connect(brave) state=start [auto_dig_mcp] 2026-08-27T16:36:55Z phase=mcp_connect(brave) state=ok [auto_dig_mcp] 2026-08-27T16:36:55Z phase=mcp_import(brave) state=start [auto_dig_mcp] 2026-08-27T16:36:55Z phase=mcp_import(brave) state=ok [auto_dig_mcp] 2026-08-27T16:36:55Z mcp_import server=brave imported_tools=8 [auto_dig_mcp] 2026-08-27T16:36:55Z phase=mcp_run(fetch) state=start [auto_dig_mcp] 2026-08-27T16:36:55Z phase=mcp_run(fetch) state=ok [auto_dig_mcp] 2026-08-27T16:36:55Z phase=mcp_connect(fetch) state=start [auto_dig_mcp] 2026-08-27T16:37:01Z phase=mcp_connect(fetch) state=ok [auto_dig_mcp] 2026-08-27T16:37:01Z phase=mcp_import(fetch) state=start [auto_dig_mcp] 2026-08-27T16:37:01Z phase=mcp_import(fetch) state=ok [auto_dig_mcp] 2026-08-27T16:37:01Z mcp_import server=fetch imported_tools=6 [auto_dig_mcp] 2026-08-27T16:37:01Z phase=capability_projection state=start [auto_dig_mcp] 2026-08-27T16:37:01Z phase=capability_projection state=ok [auto_dig_mcp] 2026-08-27T16:37:01Z capability_projection count=8 [auto_dig_mcp] 2026-08-27T16:37:01Z session_open state=ok [auto_dig_rlm] 2026-08-27T16:37:01Z phase=mcp_ready capability_count=8 capabilities=[tool('mcp.brave.brave_news_search'),tool('mcp.brave.brave_video_search'),tool('mcp.brave.brave_web_search'),tool('mcp.fetch.fetch_json'),tool('mcp.fetch.fetch_markdown'),tool('mcp.fetch.fetch_readable'),tool('mcp.fetch.fetch_txt'),tool('mcp.fetch.fetch_youtube_transcript')] [auto_dig_rlm] 2026-08-27T16:37:01Z phase=rlm_start token_budget=50000 model_calls=6 tool_calls=8 recursion_depth=2 time_limit=90.0 [auto_dig_rlm] 2026-08-27T16:37:49Z phase=rlm_return ::error title=Auto-Dig Prolog-RLM failure::status=error phase=planner kind=plan_parse_failed message=planner responses did not contain a valid typed plan used=unknown limit=unknown model_calls=3 prompt_tokens=8121 completion_tokens=4307 total_tokens=12428 cost_usd=0.00595436 [auto_dig_rlm] 2026-08-27T16:37:49Z phase=outcome status=error phase=planner kind=plan_parse_failed message=planner responses did not contain a valid typed plan used=unknown limit=unknown model_calls=3 prompt_tokens=8121 completion_tokens=4307 total_tokens=12428 cost_usd=0.00595436 [auto_dig_rlm] 2026-08-27T16:37:49Z phase=result_written file=agent-runs/auto-dig-prolog/2026-08-27-33094075463-1/rlm-result.json [auto_dig_rlm] 2026-08-27T16:37:49Z phase=trace_written file=agent-runs/auto-dig-prolog/2026-08-27-33094075463-1/rlm-trace.json [auto_dig_rlm] 2026-08-27T16:37:49Z phase=finish exit_code=1 ``` > Diagnostics are sanitized at emission time. Raw environment dumps and raw provider payloads are never included in issue-bound output.

lost-rob0t commented

2026-08-27 17:09:16 +00:00

(Migrated from github.com)

Planner-context repair is implemented and pushed on rage/277-planner-protocol-context at 4d2ab5c.

The slice adds exact typed context and tool envelopes to the mandatory root-planner protocol and makes parse/structural retries receive one bounded latest-error diagnostic. Candidate/provider output is not echoed; capability and budget denials remain non-repairable.

Verification on the exact tree:

focused completion: 41/41
focused structural retry: 3/3
deterministic PlUnit corpus: 996/996
deterministic benchmark: 16/16
deep experiment: 15/15
runtime/load checks, JSON CLI demo, and git diff --check: pass

Keep #277 open until review/merge and credentialed downstream acceptance. The separate task-framing plus retrieved-evidence composition gap will be tracked independently.

Planner-context repair is implemented and pushed on [`rage/277-planner-protocol-context`](https://github.com/lost-rob0t/prolog-rlm/tree/rage/277-planner-protocol-context) at `4d2ab5c`. The slice adds exact typed `context` and `tool` envelopes to the mandatory root-planner protocol and makes parse/structural retries receive one bounded latest-error diagnostic. Candidate/provider output is not echoed; capability and budget denials remain non-repairable. Verification on the exact tree: - focused completion: 41/41 - focused structural retry: 3/3 - deterministic PlUnit corpus: 996/996 - deterministic benchmark: 16/16 - deep experiment: 15/15 - runtime/load checks, JSON CLI demo, and `git diff --check`: pass Keep #277 open until review/merge and credentialed downstream acceptance. The separate task-framing plus retrieved-evidence composition gap will be tracked independently.

nsaspy referenced this issue

2026-08-28 03:22:03 +00:00

fix: harden planner context protocol #278

nsaspy referenced this issue

2026-08-28 03:22:34 +00:00

fix: harden planner context protocol #278

nsaspy referenced this issue

2026-09-04 22:14:07 +00:00

feat: unify task-first native model runtime #380

nsaspy closed this issue

2026-09-08 01:29:00 +00:00

nsaspy commented

2026-09-08 01:41:18 +00:00

Owner

Closing as superseded/completed by merged PR #286, which explicitly consolidated the #277 planner repair and moved the affected workload onto the unified task-first native/direct-or-plan runtime. Historical run-specific duplicates of this old planner failure are being closed in the same audit; any current regression should be represented by a fresh current-head issue.

Closing as **superseded/completed by merged PR #286**, which explicitly consolidated the #277 planner repair and moved the affected workload onto the unified task-first native/direct-or-plan runtime. Historical run-specific duplicates of this old planner failure are being closed in the same audit; any current regression should be represented by a fresh current-head issue.

nsaspy referenced this issue

2026-09-08 01:43:59 +00:00

[P0][feature-freeze] Run API/expert hardening queue with Auto-Dig dogfood repair loop #300