fix: count caps, trim-by-date, upload-asset limit, & timeout

[Copilot]

Eight correctness bugs across agentic-optimization-kit.md, copilot-token-optimizer.md, and agentic-observability-kit.md — silent data truncation, a cooldown window that breaks under load, and a Monday schedule collision causing concurrent MCP rate-limit pressure.

agentic-optimization-kit.md

• timeout 35 → 50 min
• upload-asset declared (max: 5, allowed-exts: [.png]); Phase 4 uploads 5 charts but gateway limit was absent — silent drop on the 5th
• Phase 3 count 400 → 1000; 400 runs exhaust in ~6 days on an active repo
• Hardcoded example in optimization-log entry replaced with <TODAY> / <WORKFLOW_NAME> / <TOTAL_TOKENS> / etc. placeholders — the literal 2026-04-28 / daily-observability-report was today's date and copyable verbatim
• Trim by date not count: "keep only last 30 entries" → keep entries newer than 30 days; count-based trim silently evicts entries still inside the 14-day cooldown window
• Phase timing checkpoints — date logging at phase start/end to get empirical wall-clock data before any future timeout tuning
• Jaccard — "Jaccard or cosine similarity" → "Jaccard similarity on token sets" for workflow_overlap_score
• gh api example added for reading workflow source files:

  gh api repos/{owner}/{repo}/contents/.github/workflows/WORKFLOW_NAME.md \
    --jq '.content' | base64 -d

copilot-token-optimizer.md

• jq null guard — map(.tokens) | add returns null on empty input; added // 0 after every add in the pre-aggregation script (AOK already had this fix)
• Fetch count -c 50 → -c 200; 50 runs covers ~3 days, biasing target selection
• Trim by date — same 30-day date-window fix as AOK

agentic-observability-kit.md

• Schedule weekly on monday → weekly on tuesday — eliminates the concurrent Monday MCP logs call collision with agentic-optimization-kit at zero structural cost

[Copilot]

Pull request overview

Fixes multiple correctness and operational issues in the Agentic Optimization/Observability kits and the Copilot token optimizer workflows (null-safe jq aggregation, higher run caps, safer examples/log retention guidance, asset upload limits, timeout tuning, and schedule deconfliction).

Changes:

• Increased gh aw logs fetch limits and made jq aggregations null-safe (add // 0) for Copilot token optimizer.
• Updated Agentic Optimization Kit guardrails: timeout, upload-asset constraints, larger Phase 3 count, placeholder-based examples, date-window trimming guidance, and added phase timing checkpoints + workflow-source fetch example.
• Moved Agentic Observability Kit weekly schedule from Monday to Tuesday to avoid collisions.

Show a summary per file

File	Description
.github/workflows/copilot-token-optimizer.md	Raises audit run cap, hardens jq aggregation, updates optimization-log retention guidance
.github/workflows/copilot-token-optimizer.lock.yml	Regenerated workflow lock with updated log cap and jq aggregation changes
.github/workflows/agentic-optimization-kit.md	Adds upload-asset limit + timeout increase, doc/guardrail refinements, phase timing guidance, placeholders
.github/workflows/agentic-optimization-kit.lock.yml	Regenerated lock reflecting timeout and upload-asset restriction changes
.github/workflows/agentic-observability-kit.md	Shifts weekly schedule to Tuesday
.github/workflows/agentic-observability-kit.lock.yml	Regenerated lock reflecting new Tuesday cron schedule

Copilot's findings

Tip

Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

• Files reviewed: 6/6 changed files
• Comments generated: 4

[Copilot]

The gh api repos/{owner}/{repo}/... example won’t work when copy/pasted unless {owner} and {repo} are also replaced. Consider either explicitly calling out that all placeholders must be replaced, or use an example that derives the repo from context (e.g., repos/$GITHUB_REPOSITORY/...) so only WORKFLOW_NAME needs editing.

Suggested change

	To fetch a workflow file, run (replacing `WORKFLOW_NAME` with the actual filename without extension):
	```bash
	gh api repos/{owner}/{repo}/contents/.github/workflows/WORKFLOW_NAME.md \
	To fetch a workflow file, run (replacing `WORKFLOW_NAME` with the actual filename without extension; the repository is derived from `$GITHUB_REPOSITORY`):
	```bash
	gh api "repos/$GITHUB_REPOSITORY/contents/.github/workflows/WORKFLOW_NAME.md" \

[Copilot]

jq's group_by(.workflow_name) requires the input array to be sorted by the grouping key; otherwise runs with the same workflow name that are non-adjacent will end up in different groups and totals/averages will be wrong. Add a sort_by(.workflow_name) (or equivalent) immediately before group_by(.workflow_name) in this pipeline (as done in agentic-optimization-kit.md).

See below for a potential fix:

          ]
          | sort_by(.workflow_name)

[Copilot]

The pre-aggregation jq snippet uses group_by(.workflow_name) without first sorting by workflow_name. In jq, group_by only groups adjacent items, so totals can be incorrect unless the array is sorted by the same key first. Insert | sort_by(.workflow_name) before | group_by(.workflow_name) inside this jq program.

Suggested change

run: "set -euo pipefail\nmkdir -p /tmp/gh-aw/token-audit\n\njq '{\n generated_at: (now | todateiso8601),\n window_days: 7,\n top_workflows: (\n [.runs[]\n | select(.status == \"completed\")\n | {\n workflow_name: .workflow_name,\n tokens: (.token_usage // 0),\n cost: (.estimated_cost // 0),\n turns: (.turns // 0),\n action_minutes: (.action_minutes // 0)\n }\n ]\n | group_by(.workflow_name)\n | map({\n workflow_name: .[0].workflow_name,\n run_count: length,\n total_tokens: (map(.tokens) | add // 0),\n avg_tokens: ((map(.tokens) | add // 0) / length),\n total_cost: (map(.cost) | add // 0),\n total_turns: (map(.turns) | add // 0),\n total_action_minutes: (map(.action_minutes) | add // 0)\n })\n | sort_by(.total_tokens)\n | reverse\n | .[:10]\n )\n}' /tmp/gh-aw/token-audit/all-runs.json > /tmp/gh-aw/token-audit/top-workflows.json\n\necho \"✅ Generated top workflow summary at /tmp/gh-aw/token-audit/top-workflows.json\"\njq '.top_workflows' /tmp/gh-aw/token-audit/top-workflows.json\n"

run: "set -euo pipefail\nmkdir -p /tmp/gh-aw/token-audit\n\njq '{\n generated_at: (now | todateiso8601),\n window_days: 7,\n top_workflows: (\n [.runs[]\n | select(.status == \"completed\")\n | {\n workflow_name: .workflow_name,\n tokens: (.token_usage // 0),\n cost: (.estimated_cost // 0),\n turns: (.turns // 0),\n action_minutes: (.action_minutes // 0)\n }\n ]\n | sort_by(.workflow_name)\n | group_by(.workflow_name)\n | map({\n workflow_name: .[0].workflow_name,\n run_count: length,\n total_tokens: (map(.tokens) | add // 0),\n avg_tokens: ((map(.tokens) | add // 0) / length),\n total_cost: (map(.cost) | add // 0),\n total_turns: (map(.turns) | add // 0),\n total_action_minutes: (map(.action_minutes) | add // 0)\n })\n | sort_by(.total_tokens)\n | reverse\n | .[:10]\n )\n}' /tmp/gh-aw/token-audit/all-runs.json > /tmp/gh-aw/token-audit/top-workflows.json\n\necho \"✅ Generated top workflow summary at /tmp/gh-aw/token-audit/top-workflows.json\"\njq '.top_workflows' /tmp/gh-aw/token-audit/top-workflows.json\n"

[Copilot]

This section says to "log elapsed time" at phase end, but the suggested command logs only another timestamp (no duration). Either adjust the wording to “log a timestamp” or change the guidance to actually compute and print elapsed time (e.g., capture a start epoch/SECONDS and subtract at phase end).

Suggested change

	At the start of every phase, log a timestamp with `echo "⏱️ Phase N start: $(date -u +%H:%M:%SZ)"`. At the end of each phase, log elapsed time with `echo "✅ Phase N complete: $(date -u +%H:%M:%SZ)"`. This helps identify which phases consume the most wall-clock time so future timeouts can be sized accurately.
	At the start of every phase, log a timestamp with `echo "⏱️ Phase N start: $(date -u +%H:%M:%SZ)"`. At the end of each phase, log an end timestamp with `echo "✅ Phase N complete: $(date -u +%H:%M:%SZ)"`. These checkpoints help identify which phases consume the most wall-clock time by comparing start and end times so future timeouts can be sized accurately.