From 72a33db6483c60dd8ab07d5046a1d73257ce4d65 Mon Sep 17 00:00:00 2001 From: "Founders' are 'found' to implement." Date: Tue, 15 Sep 2026 23:08:38 -0700 Subject: [PATCH 1/2] ops: refresh production anchors to 6df9b66 after #547 Extract-only skill/inventory refresh. Record #547 merge, #545 hygiene HOLD, and issue_comment skipped-run storms as reviewer-noise / not_executed. Agent-Identity: Grok (Administrator) --- .agents/skills/adaptive-wait/SKILL.md | 5 ++++- docs/ops/SKILLS-INVENTORY.md | 14 ++++++++++---- docs/ops/skills/adaptive-wait/SKILL.md | 5 ++++- 3 files changed, 18 insertions(+), 6 deletions(-) diff --git a/.agents/skills/adaptive-wait/SKILL.md b/.agents/skills/adaptive-wait/SKILL.md index e1cc66b03..69cc42cf7 100644 --- a/.agents/skills/adaptive-wait/SKILL.md +++ b/.agents/skills/adaptive-wait/SKILL.md @@ -34,6 +34,7 @@ WAIT is a **methodological stage**, not sleep-only idle time. After any action t 4. **Dual-gate before promote:** `agentic termux smoke` + `hygiene + portability gate` (or repo_gate + termux_smoke). Mergeable_state alone is insufficient. 5. **Preserve provenance.** Every attempt stays addressable by SHA. Promotion selects a successor; it does not erase failed attempts. 6. **Stall is a classification, not auto-retry permission.** Record stall class before any retry policy acts. +7. **Skipped listeners are not gates.** `issue_comment` workflows that complete `skipped` on master do not substitute for dual-gate evidence. ## Stall classes @@ -45,6 +46,7 @@ WAIT is a **methodological stage**, not sleep-only idle time. After any action t | **effect** | Workflow terminal success but expected effect (commit, artifact, page advance) missing | | **pagination** | Continuation run completes but `next_start_page` does not advance | | **routing loop** | Same SHA/input repeatedly cancels/fails without new diagnosis | +| **comment-storm-skip** | Burst of `issue_comment` runs on master with `conclusion=skipped` (Gemini/DeepSeek/ECC/Jules). Classify `not_executed` + `reviewer_noise`. Do not treat as gate failure or gate success. | ## Procedure (every wait cohort) @@ -65,6 +67,7 @@ WAIT is a **methodological stage**, not sleep-only idle time. After any action t bash sleep 25–45s # adaptive, not fixed dogma github___pull_request_read method=get_check_runs # require agentic termux smoke success + hygiene + portability gate success +# ignore skipped issue_comment listeners when scoring gates # then merge or hold ``` @@ -87,7 +90,7 @@ Promote (squash-merge) only when: - Dual gates **success** - Scope is extract-clean (or intentional docs/ops slice) -- Task outcome verified (not merely HTTP 200 / workflow green) +- Task outcome verified (not merely HTTP 200 / workflow green / skipped listener) Otherwise: HOLD, extract on fresh master branch, or CLASSIFY stall. diff --git a/docs/ops/SKILLS-INVENTORY.md b/docs/ops/SKILLS-INVENTORY.md index b762c72b1..ccc82e7fd 100644 --- a/docs/ops/SKILLS-INVENTORY.md +++ b/docs/ops/SKILLS-INVENTORY.md @@ -1,6 +1,6 @@ # Skills Inventory (termux-monorepo) -**Version:** 2026-09-16 · **Last refreshed SHA:** `97c66653` (#546) +**Version:** 2026-09-16 · **Last refreshed SHA:** `6df9b66a` (#547) **Primary agent entry:** [`CLAUDE.md`](../../CLAUDE.md) **Ops loop trigger words:** `continue`, `BIUDL`, `maximize actions`, `/continue` @@ -34,10 +34,11 @@ 1. After any dispatch/commit/merge request: **WAIT** is a stage, not idle sleep. 2. Re-check run/job state and timestamps; do concurrent non-conflicting work. -3. Stall classes: admission / queue / execution / effect / pagination / routing loop. +3. Stall classes: admission / queue / execution / effect / pagination / routing loop / **comment-storm-skip**. 4. Do not classify `queued`/`in_progress` as success or failure. -5. Promote only when dual gates green **and** task outcome verified. -6. Preserve every attempt (provenance); never rewrite history to linearize. +5. Do not treat `issue_comment` listener **skipped** conclusions as gate results. +6. Promote only when dual gates green **and** task outcome verified. +7. Preserve every attempt (provenance); never rewrite history to linearize. Local Grok project mirrors (this chat environment): - `/home/workdir/.grok/skills/evidence-led-monorepo-ops/SKILL.md` @@ -49,6 +50,8 @@ Local Grok project mirrors (this chat environment): Minimum activity taxonomy: `actionable_finding`, `provider_state`, `reviewer_noise`, `execution_failure`, `not_executed`, `self_trigger_candidate`. ECC-tools/Codex/Qodo quota, billing, availability, permission, and status chatter is provider-state/reviewer-noise unless bound execution evidence demonstrates an actual task failure. +`issue_comment` storms that complete as **skipped** on master (Gemini/DeepSeek/ECC/Jules dispatch) are `not_executed` + `reviewer_noise`, not `execution_failure`. + ## Supporting skills (repo) | Skill | Path | @@ -72,6 +75,7 @@ Minimum activity taxonomy: `actionable_finding`, `provider_state`, `reviewer_noi - **#537**: README fast-start leads with CLAUDE.md; skills inventory + deploy lanes linked. - **#542**: Linguist CedrLang document short-circuit + fence guard. - **#546**: reviewer-noise taxonomy + production anchors on master. +- **#547**: skill anchors refreshed to `97c66653` post-#546; now master is `6df9b66a`. ## Backfill / stall context (ops) @@ -81,6 +85,8 @@ Minimum activity taxonomy: `actionable_finding`, `provider_state`, `reviewer_noi - **#523** / **#527** remain **HOLD** (mega). - **#544** superseded by **#546** (do not wholesale-merge). - **#543** HOLD for skill-quality evaluator extract; `validate-pull-request` still red on stale base. +- **#545** HOLD: `agentic termux smoke` success, `hygiene + portability gate` failure; behind master (`5134b6a7`). +- **#455** / **#48** comment activity fans out skipped listeners on master — observe, do not merge wholesale. ## Cross-session rule diff --git a/docs/ops/skills/adaptive-wait/SKILL.md b/docs/ops/skills/adaptive-wait/SKILL.md index e1cc66b03..69cc42cf7 100644 --- a/docs/ops/skills/adaptive-wait/SKILL.md +++ b/docs/ops/skills/adaptive-wait/SKILL.md @@ -34,6 +34,7 @@ WAIT is a **methodological stage**, not sleep-only idle time. After any action t 4. **Dual-gate before promote:** `agentic termux smoke` + `hygiene + portability gate` (or repo_gate + termux_smoke). Mergeable_state alone is insufficient. 5. **Preserve provenance.** Every attempt stays addressable by SHA. Promotion selects a successor; it does not erase failed attempts. 6. **Stall is a classification, not auto-retry permission.** Record stall class before any retry policy acts. +7. **Skipped listeners are not gates.** `issue_comment` workflows that complete `skipped` on master do not substitute for dual-gate evidence. ## Stall classes @@ -45,6 +46,7 @@ WAIT is a **methodological stage**, not sleep-only idle time. After any action t | **effect** | Workflow terminal success but expected effect (commit, artifact, page advance) missing | | **pagination** | Continuation run completes but `next_start_page` does not advance | | **routing loop** | Same SHA/input repeatedly cancels/fails without new diagnosis | +| **comment-storm-skip** | Burst of `issue_comment` runs on master with `conclusion=skipped` (Gemini/DeepSeek/ECC/Jules). Classify `not_executed` + `reviewer_noise`. Do not treat as gate failure or gate success. | ## Procedure (every wait cohort) @@ -65,6 +67,7 @@ WAIT is a **methodological stage**, not sleep-only idle time. After any action t bash sleep 25–45s # adaptive, not fixed dogma github___pull_request_read method=get_check_runs # require agentic termux smoke success + hygiene + portability gate success +# ignore skipped issue_comment listeners when scoring gates # then merge or hold ``` @@ -87,7 +90,7 @@ Promote (squash-merge) only when: - Dual gates **success** - Scope is extract-clean (or intentional docs/ops slice) -- Task outcome verified (not merely HTTP 200 / workflow green) +- Task outcome verified (not merely HTTP 200 / workflow green / skipped listener) Otherwise: HOLD, extract on fresh master branch, or CLASSIFY stall. From 0bb0ab27f0e70a7c4f018aea99ac883724cc30ac Mon Sep 17 00:00:00 2001 From: "Founders' are 'found' to implement." Date: Tue, 15 Sep 2026 23:09:09 -0700 Subject: [PATCH 2/2] ops: update evidence-led anchors after #547 merge Master HEAD 6df9b66a. HOLD #545 hygiene-red. comment-storm-skip taxonomy. Agent-Identity: Grok (Administrator) --- .../skills/evidence-led-monorepo-ops/SKILL.md | 19 +++++++++++-------- .../skills/evidence-led-monorepo-ops/SKILL.md | 19 +++++++++++-------- 2 files changed, 22 insertions(+), 16 deletions(-) diff --git a/.agents/skills/evidence-led-monorepo-ops/SKILL.md b/.agents/skills/evidence-led-monorepo-ops/SKILL.md index d45f6343d..8712bea53 100644 --- a/.agents/skills/evidence-led-monorepo-ops/SKILL.md +++ b/.agents/skills/evidence-led-monorepo-ops/SKILL.md @@ -40,6 +40,7 @@ description: Continuous evidence-led admin ops on timerloggedout-spec/termux-mon - Candidate PRs: `get` + `get_check_runs` - Require dual-gate success before merge - Dirty / conflicted / behind-master with extra failures → **HOLD** or **extract** on fresh branch from master + - Ignore `issue_comment` skipped listeners when scoring gates 3. **Triage** - Prefer: security/perf 1–5 file extracts, docs stubs, admission fixes, skill-anchor refresh @@ -53,7 +54,7 @@ description: Continuous evidence-led admin ops on timerloggedout-spec/termux-mon 5. **Adaptive WAIT** - Align with `adaptive-feedback-cycle` + `production-reconciliation` - - Stall classes: admission / queue / execution / effect / pagination / routing loop + - Stall classes: admission / queue / execution / effect / pagination / routing loop / comment-storm-skip - Multi-pass: re-poll after disposition before closing cycle 6. **Feed forward** @@ -62,18 +63,18 @@ description: Continuous evidence-led admin ops on timerloggedout-spec/termux-mon ## Current production anchors (refresh on each cycle) -| Item | State (2026-09-16T04:14Z) | +| Item | State (2026-09-16T06:08Z) | |------|--------------------| -| Master HEAD | `97c66653` (#546 reviewer-noise + anchors; prior `5134b6a7` #542/#541/#524) | -| Codespace agent lane | #530 + #531 MERGED; #545 multi-lane configs open | +| Master HEAD | `6df9b66a` (#547 anchor refresh; prior `97c66653` #546, `5134b6a7` #542/#541/#524) | +| Codespace agent lane | #530 + #531 MERGED; #545 multi-lane **HOLD** (hygiene red, smoke green, base `5134b6a7`) | | AGENTS→CLAUDE fold | #488 + #534; CLAUDE.md primary | | Skills inventory | `docs/ops/SKILLS-INVENTORY.md` | | Backfill admission | schedule `23 * * * *` on `context-relationship-backfill.yml`; default page **2** (stalled since 2026-08-19) | | #526 audit | Landed; Tanka abandoned temporary quota — findings valid | | HOLD mega | #523, #527, #142, #455 conflicted, staging #48 | -| Behind-master / extract | #543 skill-quality HOLD (dual-gate green, `validate-pull-request` red, base stale); #544 superseded by #546 | -| Dual gates on `97c66653` | `termux smoke` success + `repo gate` success | -| Immediate-fail master runs | `swe-reference-evaluation`, `historical-evaluation-correlation`, `agent-jules-on-issues`, `actions-run-watcher` — classify `not_executed` until job/step logs prove execution | +| Behind-master / extract | #543 skill-quality HOLD; #544 superseded by #546; #545 hygiene HOLD | +| Dual gates on `6df9b66a` | Prior extract dual-gate green on `954e2022` before squash; do not use skipped `issue_comment` runs as gate evidence | +| Immediate-fail / skip master runs | `swe-reference-evaluation`, `historical-evaluation-correlation`, `agent-jules-on-issues`, `actions-run-watcher` immediate-fail → `not_executed` until logs prove execution; Gemini/DeepSeek/ECC/Jules `issue_comment` **skipped** → `comment-storm-skip` | ## P0 classification @@ -83,6 +84,7 @@ description: Continuous evidence-led admin ops on timerloggedout-spec/termux-mon | gate-failure | smoke/hygiene red | fix extract or hold | | admission-stall | 0 runs on scheduled workflow | add schedule or dispatch (done for backfill) | | comment-loop / self-trigger | bot↔bot issue_comment storms | observe; Tier-4 workflow fix | +| comment-storm-skip | many skipped listeners on master SHA | `not_executed` + `reviewer_noise`; do not merge or revert on that signal | | reviewer-noise | quota/billing/availability/status chatter without a task failure | classify as provider-state telemetry; do not promote to code failure | | security | path traversal, Dependabot | extract fix + dual-gate | | dirty-mega | 20+ files / thousands of lines | extract-only | @@ -97,7 +99,7 @@ For every retrospective slice, classify automation-generated activity before int - `provider_state`: quota, billing, unavailable integration, permission, or rate-limit state. - `reviewer_noise`: informational/status output that does not establish a repository defect or failed task. - `execution_failure`: a task/run actually started and failed, with bound run/step evidence. -- `not_executed`: admission, availability, cooldown, quota, or routing prevented execution; **never relabel this as `execution_failure`**. +- `not_executed`: admission, availability, cooldown, quota, routing, or **skipped listener** prevented execution; **never relabel this as `execution_failure`**. - `self_trigger_candidate`: a reactive listener plus a matching self-posting path; do not flag read-only listeners merely because they consume comments. When evaluating #390-like behavior, correlate issue comments, review submissions/comments, workflow runs, jobs/steps, artifacts, commits, PRs, and actor/provider state by immutable IDs plus SHA/ref and timestamps. Detect causal loops without allowing reviewer/provider chatter to contaminate effectiveness metrics. @@ -126,6 +128,7 @@ Retroactive review is **continuous**, with periodic sampling as a backstop; it m - Sleep-only WAIT with no concurrent useful work - Acting on quota_cooldown / provider-control as if real_review - Treating ECC-tools/Codex/Qodo status chatter as independent task failures without execution evidence +- Treating skipped `issue_comment` listener runs as dual-gate results - Duplicating skill content only in chat — **always commit to master** ## Skill maintenance diff --git a/docs/ops/skills/evidence-led-monorepo-ops/SKILL.md b/docs/ops/skills/evidence-led-monorepo-ops/SKILL.md index d45f6343d..8712bea53 100644 --- a/docs/ops/skills/evidence-led-monorepo-ops/SKILL.md +++ b/docs/ops/skills/evidence-led-monorepo-ops/SKILL.md @@ -40,6 +40,7 @@ description: Continuous evidence-led admin ops on timerloggedout-spec/termux-mon - Candidate PRs: `get` + `get_check_runs` - Require dual-gate success before merge - Dirty / conflicted / behind-master with extra failures → **HOLD** or **extract** on fresh branch from master + - Ignore `issue_comment` skipped listeners when scoring gates 3. **Triage** - Prefer: security/perf 1–5 file extracts, docs stubs, admission fixes, skill-anchor refresh @@ -53,7 +54,7 @@ description: Continuous evidence-led admin ops on timerloggedout-spec/termux-mon 5. **Adaptive WAIT** - Align with `adaptive-feedback-cycle` + `production-reconciliation` - - Stall classes: admission / queue / execution / effect / pagination / routing loop + - Stall classes: admission / queue / execution / effect / pagination / routing loop / comment-storm-skip - Multi-pass: re-poll after disposition before closing cycle 6. **Feed forward** @@ -62,18 +63,18 @@ description: Continuous evidence-led admin ops on timerloggedout-spec/termux-mon ## Current production anchors (refresh on each cycle) -| Item | State (2026-09-16T04:14Z) | +| Item | State (2026-09-16T06:08Z) | |------|--------------------| -| Master HEAD | `97c66653` (#546 reviewer-noise + anchors; prior `5134b6a7` #542/#541/#524) | -| Codespace agent lane | #530 + #531 MERGED; #545 multi-lane configs open | +| Master HEAD | `6df9b66a` (#547 anchor refresh; prior `97c66653` #546, `5134b6a7` #542/#541/#524) | +| Codespace agent lane | #530 + #531 MERGED; #545 multi-lane **HOLD** (hygiene red, smoke green, base `5134b6a7`) | | AGENTS→CLAUDE fold | #488 + #534; CLAUDE.md primary | | Skills inventory | `docs/ops/SKILLS-INVENTORY.md` | | Backfill admission | schedule `23 * * * *` on `context-relationship-backfill.yml`; default page **2** (stalled since 2026-08-19) | | #526 audit | Landed; Tanka abandoned temporary quota — findings valid | | HOLD mega | #523, #527, #142, #455 conflicted, staging #48 | -| Behind-master / extract | #543 skill-quality HOLD (dual-gate green, `validate-pull-request` red, base stale); #544 superseded by #546 | -| Dual gates on `97c66653` | `termux smoke` success + `repo gate` success | -| Immediate-fail master runs | `swe-reference-evaluation`, `historical-evaluation-correlation`, `agent-jules-on-issues`, `actions-run-watcher` — classify `not_executed` until job/step logs prove execution | +| Behind-master / extract | #543 skill-quality HOLD; #544 superseded by #546; #545 hygiene HOLD | +| Dual gates on `6df9b66a` | Prior extract dual-gate green on `954e2022` before squash; do not use skipped `issue_comment` runs as gate evidence | +| Immediate-fail / skip master runs | `swe-reference-evaluation`, `historical-evaluation-correlation`, `agent-jules-on-issues`, `actions-run-watcher` immediate-fail → `not_executed` until logs prove execution; Gemini/DeepSeek/ECC/Jules `issue_comment` **skipped** → `comment-storm-skip` | ## P0 classification @@ -83,6 +84,7 @@ description: Continuous evidence-led admin ops on timerloggedout-spec/termux-mon | gate-failure | smoke/hygiene red | fix extract or hold | | admission-stall | 0 runs on scheduled workflow | add schedule or dispatch (done for backfill) | | comment-loop / self-trigger | bot↔bot issue_comment storms | observe; Tier-4 workflow fix | +| comment-storm-skip | many skipped listeners on master SHA | `not_executed` + `reviewer_noise`; do not merge or revert on that signal | | reviewer-noise | quota/billing/availability/status chatter without a task failure | classify as provider-state telemetry; do not promote to code failure | | security | path traversal, Dependabot | extract fix + dual-gate | | dirty-mega | 20+ files / thousands of lines | extract-only | @@ -97,7 +99,7 @@ For every retrospective slice, classify automation-generated activity before int - `provider_state`: quota, billing, unavailable integration, permission, or rate-limit state. - `reviewer_noise`: informational/status output that does not establish a repository defect or failed task. - `execution_failure`: a task/run actually started and failed, with bound run/step evidence. -- `not_executed`: admission, availability, cooldown, quota, or routing prevented execution; **never relabel this as `execution_failure`**. +- `not_executed`: admission, availability, cooldown, quota, routing, or **skipped listener** prevented execution; **never relabel this as `execution_failure`**. - `self_trigger_candidate`: a reactive listener plus a matching self-posting path; do not flag read-only listeners merely because they consume comments. When evaluating #390-like behavior, correlate issue comments, review submissions/comments, workflow runs, jobs/steps, artifacts, commits, PRs, and actor/provider state by immutable IDs plus SHA/ref and timestamps. Detect causal loops without allowing reviewer/provider chatter to contaminate effectiveness metrics. @@ -126,6 +128,7 @@ Retroactive review is **continuous**, with periodic sampling as a backstop; it m - Sleep-only WAIT with no concurrent useful work - Acting on quota_cooldown / provider-control as if real_review - Treating ECC-tools/Codex/Qodo status chatter as independent task failures without execution evidence +- Treating skipped `issue_comment` listener runs as dual-gate results - Duplicating skill content only in chat — **always commit to master** ## Skill maintenance