implement SKILL.md references a missing escalation-procedure.md; suite-check and test-writer often return without next_step #605

New issue

Closed

opened 2026-09-16 21:27:49 +00:00 by jared · 1 comment

jared commented

2026-09-16 21:27:49 +00:00

Owner

Context

plugins/os-sdlc/skills/implement/SKILL.md says read and follow the escalation procedure: references/escalation-procedure.md but the installed skill has no references/ directory (checked 2026-09-16 in ~/.claude/plugins/os-sdlc/skills/implement/). On escalation the orchestrator has no procedure to follow. Separately, in impl 23 and 24 the programmer, both suite-checks, and three test-writer dispatches returned Done. with no next_step JSON; the orchestrator recovered each time with os-sdlc-runner next. suite-check dispatches took 733 s wall against a 280 s suite step: about 450 s of agent overhead per suite-check, twice per run.

Observed

Escalation reference missing. Agent dispatches return no next_step JSON. suite-check overhead is ~450 s per run.

Reproduce

n/a: design defects in implement SKILL.md and agent charters.

Expected

The escalation reference file exists and the SKILL.md pointer resolves. A suite-check dispatch final message contains the next_step JSON (charter test or validator). Timing instrumentation shows where suite-check spends its non-suite 450 s.

Origin

Trigger: impl 23 and 24 orchestration, session 01153d94
Improvised this session: none
Chain: implement SKILL.md points to missing file → orchestrator blocked on escalation DESIGN
Root candidate: suite-check and agent charter instrumentation
Where: plugins/os-sdlc/skills/implement/SKILL.md, plugins/os-sdlc/agents/suite-check.md
Session: 01153d94-6459-43dc-bd20-e68b757af56f
Transcript: https://claude.com/claude-code (session in progress)

Owner sketch

class OsSdlc::Skills::Implement
  def escalation_procedure_path
end

class OsSdlc::Agents::SuiteCheck
  def run(ticket_brief)
end

Cases

The escalation reference file exists at references/escalation-procedure.md and the SKILL.md pointer resolves.
A suite-check dispatch final message contains the subagent-stop next_step JSON verbatim (charter test or validator).
Timing instrumentation shows where suite-check spends its non-suite 450 s overhead per run.

## Context plugins/os-sdlc/skills/implement/SKILL.md says read and follow the escalation procedure: references/escalation-procedure.md but the installed skill has no references/ directory (checked 2026-09-16 in ~/.claude/plugins/os-sdlc/skills/implement/). On escalation the orchestrator has no procedure to follow. Separately, in impl 23 and 24 the programmer, both suite-checks, and three test-writer dispatches returned Done. with no next_step JSON; the orchestrator recovered each time with `os-sdlc-runner next`. suite-check dispatches took 733 s wall against a 280 s suite step: about 450 s of agent overhead per suite-check, twice per run. ## Observed Escalation reference missing. Agent dispatches return no next_step JSON. suite-check overhead is ~450 s per run. ## Reproduce n/a: design defects in implement SKILL.md and agent charters. ## Expected The escalation reference file exists and the SKILL.md pointer resolves. A suite-check dispatch final message contains the next_step JSON (charter test or validator). Timing instrumentation shows where suite-check spends its non-suite 450 s. ## Origin - Trigger: impl 23 and 24 orchestration, session 01153d94 - Improvised this session: none - Chain: implement SKILL.md points to missing file → orchestrator blocked on escalation DESIGN - Root candidate: suite-check and agent charter instrumentation - Where: plugins/os-sdlc/skills/implement/SKILL.md, plugins/os-sdlc/agents/suite-check.md - Session: 01153d94-6459-43dc-bd20-e68b757af56f - Transcript: https://claude.com/claude-code (session in progress) ## Owner sketch ```ruby class OsSdlc::Skills::Implement def escalation_procedure_path end class OsSdlc::Agents::SuiteCheck def run(ticket_brief) end ``` ## Cases 1. The escalation reference file exists at references/escalation-procedure.md and the SKILL.md pointer resolves. 2. A suite-check dispatch final message contains the subagent-stop next_step JSON verbatim (charter test or validator). 3. Timing instrumentation shows where suite-check spends its non-suite 450 s overhead per run.

jared added the

plugin/os-sdlc

filed-by/agent

labels

2026-09-16 21:27:49 +00:00

jared referenced this issue

2026-09-17 07:32:53 +00:00

lsp-probe symbols falls back to an ast-grep definition search when workspace/symbol returns nothing #603

jared referenced this issue from a commit

2026-09-22 17:10:37 +00:00

feat(os-sdlc): subagent-stop reply ends with next_step and a final_output relay instruction; charters put Final output last (#605 cases 1-2, ADR-0203)

jared commented

2026-09-22 18:31:11 +00:00

Author

Owner

Resolution

Done: Case 1: escalation-procedure reference fixed in b1a10cdc (2026-09-17). Case 2: charter final-output relay moved to the end of the suite-check, test-writer, and programmer charters in e1f33d96 (ADR-0203). Case 3: the 450 s suite-check overhead did not reproduce; no instrumentation added.

Evidence: Live run: implementation 10 on #557 (session e8c75cf9, 2026-09-22) ran 8 dispatches, 0 retries, no hand settlement. test-writer (123), programmer (125), and suite-check (127) each ended with the dispatch JSON relay. suite-check elapsed 131 s: gate command ~123 s, agent overhead under 2 s. Per-dispatch wall time is already available from dispatches.created_at and completed_at.

Follow-ups: none. Observations from the transcript review (LSP daemon spawn timeout flake in the suite, suite-check echoing its dispatch JSON after settlement, programmer gate ~130 s per run) reported to the user for triage; not filed.

## Resolution **Done:** Case 1: escalation-procedure reference fixed in b1a10cdc (2026-09-17). Case 2: charter final-output relay moved to the end of the suite-check, test-writer, and programmer charters in e1f33d96 (ADR-0203). Case 3: the 450 s suite-check overhead did not reproduce; no instrumentation added. **Evidence:** Live run: implementation 10 on #557 (session e8c75cf9, 2026-09-22) ran 8 dispatches, 0 retries, no hand settlement. test-writer (123), programmer (125), and suite-check (127) each ended with the dispatch JSON relay. suite-check elapsed 131 s: gate command ~123 s, agent overhead under 2 s. Per-dispatch wall time is already available from dispatches.created_at and completed_at. **Follow-ups:** none. Observations from the transcript review (LSP daemon spawn timeout flake in the suite, suite-check echoing its dispatch JSON after settlement, programmer gate ~130 s per run) reported to the user for triage; not filed.

jared closed this issue

2026-09-22 18:31:12 +00:00

jared referenced this issue

2026-09-23 20:23:45 +00:00

Prove the map runner live: hook-driven end-to-end drive #454

jared referenced this issue

2026-09-23 20:23:48 +00:00

Finished agents re-emit next_step nine times after subagent-stop #529