Analyzing Session Work Logs with diagnosing-superpowers
機械翻訳 / Machine-translated

機械翻訳 / Machine-translated
Original article: Analyzing Session Work Logs with diagnosing-superpowers (cccskills)
Skill page: diagnosing-superpowers
Key takeaways from this article
diagnosing-superpowers is a skill for people using superpowers who are concerned about duplicated work, deviations from plans, or excessive time and token usage in their sessions. It compiles session records into a report with line number references. This article introduces the seven analytical perspectives it covers, how to install it, and example requests. Notes on using it for bug reports are also included.
With diagnosing-superpowers installed, you can receive a report summarizing what happened during a session that used superpowers. The report includes line number references from the log files. For example, you might make a request like this:
Please check the records to find out why the same test was run over and over again in the last session.
The agent doesn't start analyzing immediately — it returns questions one at a time. It confirms four things: the target session, the expected outcome, what actually happened, and the metrics you're concerned about. Metrics include elapsed time, tokens, repeated operations, and so on.
Once the problem is defined, the agent identifies the session's log file by its absolute path. For past sessions, it quotes the original request and timestamp to confirm the target. Items ruled out as candidates are also shown, along with the reasons.
Next, it creates a working folder and tells you its path:
~/.superpowers/diagnosing-superpowers/<session-ID>/
The analysis is divided into seven perspectives, each handled by a subagent (an auxiliary agent running in parallel). The report is written to this folder and also displayed on screen. Every finding includes "file path: line number," and any finding without a supporting line in the records is excluded from the report.
This skill lets you examine a session that didn't go well from seven perspectives and confirm what happened with evidence. The perspectives and what they examine are as follows:
| Perspective | What it examines |
|---|---|
| Skill usage | Which skills were used, when, and which were not |
| Plan adherence | Whether work proceeded according to the created plan |
| Work duplication | Whether the same work was repeated |
| Stumbling points | Where work got stuck |
| Quality and evidence | The quality of results and the records supporting them |
| Conflicts with requests | Actions that conflicted with the user's requests |
| Time and consumption | Elapsed time and token consumption |
All seven analyses are run every time. What changes based on how you express your complaint is which section the agent reads first and which perspective is highlighted in the conclusion.
| Example complaint | Perspective highlighted first |
|---|---|
| It took too long | Time and consumption, stumbling points |
| Why was unnecessary work done? | Work duplication, plan adherence |
| Why is this costing so much? | Time and consumption |
| I can't tell what it's doing while it runs | Skill usage (noting that it was mid-execution) |
| My plan was ignored | Plan adherence (checked from the line where conversation compression occurred) |
| A specific skill didn't work | Skill usage |
Another notable feature is the strict handling of numbers. Token counts and times are always taken from the records or from the results of commands the agent executed — never supplemented from memory or general pricing estimates.
The scope of analysis is not limited to the current session. Past sessions specified by ID or path, as well as subagent records, are also included.
There are two ways to install: via a command, or by asking the agent. For the command approach, run the following. After running it, you can choose the target agent from options such as Claude Code or Codex.
npx skills add https://github.com/obra/superpowers --skill diagnosing-superpowers
To ask the agent, send the following message as-is. The workflow involves reviewing the contents first, then placing the files.
Please install the following Agent Skill.
https://github.com/obra/superpowers/tree/main/skills/diagnosing-superpowers
First, check the contents of SKILL.md and its accompanying files to confirm there are no suspicious operations (such as external data transmission or reading of credentials), then place them in ~/.claude/skills/diagnosing-superpowers/ for Claude Code, or ~/.agents/skills/diagnosing-superpowers/ for Codex.
Installation locations are as follows. Use the user-level location to make the skill available across all projects, or the project-level location to limit it to a specific project.
| Agent | User-level | Project-level |
|---|---|---|
| Claude Code | ~/.claude/skills/ | .claude/skills/ |
| Codex | ~/.agents/skills/ | .agents/skills/ |
The distribution source is obra/superpowers. Since this skill reads only the session logs from sessions that used superpowers, it assumes that superpowers was used in the session you want to analyze.
The clearer you are about the target session and your concern, the faster the process moves. A request that specifies both the target and what to examine is treated as a problem description in itself.
Please investigate why yesterday's session (ID: ○○) took longer than expected, focusing on time and token consumption.
The resulting report leads with findings on time and consumption and stumbling points. The time and token figures include references to which lines of the log they were taken from.
Please report on why the brainstorming skill was not used in this session, based on the records.
In this case, the report centers on skill usage. The progression of which skills were used at which point is shown with line number references.
Based on the report, please draft a GitHub issue for superpowers.
The agent first searches open and closed issues for similar symptoms. If a close match is found, it suggests adding a comment to that issue; if not, it drafts new text and shows the full content. It only posts after receiving your approval.
You can also create a shareable bundle of session records for bug reports. A bundle is only created when you ask for one. You can choose from three levels of what to include:
| Level | Contents |
|---|---|
| skeleton | Does not include the body of tool execution results |
| evidence | Includes the body only for sections quoted in the report |
| full | Includes everything |
The most important thing to know is that this skill does not assert causes or suggest fixes. The report states only whether superpowers was involved — it does not name skill defects or propose changes. No matter how strongly you push for fixes, this boundary does not change.
Other things to be aware of:
gh is used to post issues. Since gh does not support file attachments, you attach the bundle manually via your browser.Even with scrubbing, some sensitive information may slip through. Before sharing, review all files in the bundle yourself. For the report as well, it's safest to open the original log using the line numbers as a guide and verify the quoted content yourself.
The license is MIT. Before installing, check SKILL.md and the accompanying files to confirm there are no suspicious operations such as external data transmission.
Here are skills from the same obra/superpowers collection that handle the actual work targeted by diagnosing-superpowers analyses.
executing-plans is a skill that executes a completed implementation plan task by task within the same session. If you feel a session using this skill went off-plan, you can review it with diagnosing-superpowers.
subagent-driven-development is a skill that delegates plan tasks to separate agents. Since subagent records are also within scope for analysis, it can be paired with diagnosing-superpowers to verify work duplication or how time was spent.
This skill is intended for cases where a session using superpowers didn't go as expected. The reports are also structured around describing superpowers' involvement. It is not suited for reviewing work unrelated to superpowers.
No. Log files are read-only — they are never modified, moved, or deleted. Outputs such as reports are written to a dedicated working folder.
No. The agent shows the full issue text and creates it only after receiving your approval. It will never post with placeholder text still present.
No. It reports only on events that can be confirmed in the records. Identifying skill defects or proposing changes is outside its scope. Whether to change superpowers is a decision made by those who receive and triage the issues and bundles.
diagnosing-superpowers is a skill that compiles the history of sessions using superpowers into a report with line number references. It lets you verify work duplication, plan deviations, and time and token consumption from the records, and it can also help prepare bug reports. However, posting and creating shareable data both require your approval before proceeding. More details are available on the cccskills page.
This article is a reprint of the cccskills article "Analyzing Session Work Logs with diagnosing-superpowers."
Skill details and installation instructions can be found on the cccskills diagnosing-superpowers page.