🎯 Target Workflow
Release — selected as the highest-priority LLM-based workflow not excluded by name (Token-named workflows excluded) with analyzable prompt source.
- Runs analyzed: 2 (2026-05-14 and 2026-05-20)
- Analysis period: 2026-05-13 → 2026-05-20 (7 days)
- Conclusions: 2 × success (100% success rate)
- Agent job duration: ~2.5 min per run (12:52:58 → 12:55:34 on the most recent run)
📊 Token Profile
Note: Per-run telemetry data was unavailable in the pre-aggregated snapshots (all token_usage fields were 0 in all-runs.json and daily snapshots). The analysis below is derived from prompt source inspection and structural counting.
| Metric |
Value |
| Runs in period |
2 |
| Avg agent job duration |
~2.5 min |
| Prompt body size |
~8,100 chars (~2,025 tokens) |
| Prompt section with highest overhead |
## What to Write (~1,650 tokens, 82% of prompt body) |
| Inline sub-agents |
None |
| Tool configuration |
bash: ["*"], update-release |
🏆 Ranked Recommendations
1. Merge Redundant Constraint Sections — Est. ~350 tokens/run saved
Evidence: The ## What to Write section contains four distinct sub-sections that all communicate the same family of constraint: focus on user-facing impact, exclude internal details, write well:
- Selection rubric (5 bullets)
- Exclude internal-only changes (4 bullets)
- Release-note anti-patterns to avoid (6 bullets)
- Writing quality constraints (6 bullets)
These 21 bullets (~450 words) heavily overlap. For example:
- "omit it when it only affects how the repository is built, tested, compiled..." (rubric) ≈ "CI, build, test, refactor, repo maintenance..." (exclude internal)
- "rewriting commit subjects into bullets" (anti-patterns) ≈ "translate implementation changes into user impact" (user-facing section above)
- "use plain English and strong verbs" + "keep bullets parallel" (writing quality) could be a single inline note
Action: Consolidate these four sections into one ### Constraints block with ~10 bullets, removing ~11 redundant bullets.
Estimated savings: ~350 tokens per run (~17% of prompt body)
2. Condense Keep a Changelog Section Definitions — Est. ~180 tokens/run saved
Evidence: The prompt dedicates 6 bullets to defining each Keep a Changelog section type (Added, Changed, Deprecated, Removed, Fixed, Security) with detailed usage criteria. Keep a Changelog is a well-known convention; this level of definition adds overhead without meaningful accuracy gains.
Current text (excerpt):
- `Added` for new user-visible capabilities, new workflows, new supported inputs...
- `Changed` for meaningful behavior changes, revised defaults, performance changes...
- `Deprecated` only when something still works in this release but users should migrate...
Action: Replace the 6-bullet section-by-section guide with a single line:
Follow Keep a Changelog conventions (Added / Changed / Fixed / Removed / Security / Deprecated); omit any section with no strong user-facing entry.
Estimated savings: ~180 tokens per run (~9% of prompt body)
3. Trim Semver Section Duplication — Est. ~100 tokens/run saved
Evidence: The ### The notes must be semver-aware block has 7 bullets. Bullets 5–7 (for a major release... for a minor release... for a patch release...) restate guidance already given in bullet 1, creating duplication within the same section.
Action: Collapse bullets 5–7 into a single inline parenthetical appended to bullet 1:
Infer intended semantic meaning from the tag and diff, not commit prefixes (major: breaking changes and migration steps; minor: new capabilities; patch: fixes and polish).
Estimated savings: ~100 tokens per run (~5% of prompt body)
📈 Combined Impact
| Recommendation |
Est. Tokens Saved/Run |
Confidence |
| Merge constraint sections |
~350 |
High |
| Condense KaC section definitions |
~180 |
High |
| Trim semver duplication |
~100 |
Medium |
| Total |
~630 |
— |
At ~2,025 prompt tokens, this represents a ~31% reduction in prompt size with no expected loss of output quality. The Release workflow is low-frequency (~2 runs/week) but each optimization persists across all future runs indefinitely.
⚠️ Caveats
- Actual token telemetry was unavailable (all pre-aggregated data showed 0 tokens); savings estimates are based on token counts derived from prompt character counts (~4 chars/token).
- Only 2 runs were available in the analysis window; a broader sample could surface additional turn-level inefficiencies.
- The CHANGELOG update + git push behavior is complex and intentionally detailed — those instructions should not be simplified.
References:
Generated by Agentic Workflow Token Usage Optimizer · ● 9M · ◷
🎯 Target Workflow
Release — selected as the highest-priority LLM-based workflow not excluded by name (Token-named workflows excluded) with analyzable prompt source.
📊 Token Profile
## What to Write(~1,650 tokens, 82% of prompt body)bash: ["*"],update-release🏆 Ranked Recommendations
1. Merge Redundant Constraint Sections — Est. ~350 tokens/run saved
Evidence: The
## What to Writesection contains four distinct sub-sections that all communicate the same family of constraint: focus on user-facing impact, exclude internal details, write well:These 21 bullets (~450 words) heavily overlap. For example:
Action: Consolidate these four sections into one
### Constraintsblock with ~10 bullets, removing ~11 redundant bullets.Estimated savings: ~350 tokens per run (~17% of prompt body)
2. Condense Keep a Changelog Section Definitions — Est. ~180 tokens/run saved
Evidence: The prompt dedicates 6 bullets to defining each Keep a Changelog section type (Added, Changed, Deprecated, Removed, Fixed, Security) with detailed usage criteria. Keep a Changelog is a well-known convention; this level of definition adds overhead without meaningful accuracy gains.
Current text (excerpt):
Action: Replace the 6-bullet section-by-section guide with a single line:
Estimated savings: ~180 tokens per run (~9% of prompt body)
3. Trim Semver Section Duplication — Est. ~100 tokens/run saved
Evidence: The
### The notes must be semver-awareblock has 7 bullets. Bullets 5–7 (for a major release... for a minor release... for a patch release...) restate guidance already given in bullet 1, creating duplication within the same section.Action: Collapse bullets 5–7 into a single inline parenthetical appended to bullet 1:
Estimated savings: ~100 tokens per run (~5% of prompt body)
📈 Combined Impact
At ~2,025 prompt tokens, this represents a ~31% reduction in prompt size with no expected loss of output quality. The Release workflow is low-frequency (~2 runs/week) but each optimization persists across all future runs indefinitely.
References: