How to audit agent sessions across projects for platform engineers in 2026

Audit agent sessions across projects: gather the records, group them by project and account, and check usage and cost evidence.

By SupaKeeper · 2026-10-27

SupaKeeper is published by LunarWerx, which makes some of the tools compared here.

To audit agent sessions across projects, define which sessions and accounts need review, collect the available session and usage details, group findings by project, and check that the records support the questions the audit needs to answer. AgentHydra shows Claude Code, Codex and OpenCode sessions on your machine in one searchable list, updated live while they run. ^[1] Compare it with existing workflows and other session management options before choosing how to collect and review the information.

Option	What it can do	What to check
AgentHydra	Shows Claude Code, Codex and OpenCode sessions on your machine in one searchable list, updated live while they run. ^[1] Shows cost at API list prices by day, model, project and account. ^[1]	Check whether its session view, account details, cost estimates, and supported operating system match the audit requirements. ^[1] ^[1]
Existing agent workflow	Use the session views and organization methods already available in the workflow.	Check which agent types, project details, accounts, and historical records are available.
Session management tool	A separate tool may provide a way to organize session details.	Check supported sessions, available account information, historical coverage, and actions before choosing it.
Usage and billing records	Use available service records to review the usage and cost information they contain.	Check the covered period, project and account identifiers, and whether cost figures are estimates or actual recorded charges.

What you need before you start

Before auditing agent sessions across projects, define what the review needs to establish. Platform engineers may need to identify active work, connect sessions with projects and accounts, review usage, or find recurring command failures. These are separate questions, so decide which ones the audit must answer before selecting a tool.

Prepare a list of the projects, accounts, and agent session types that fall within the review. Record the available identifiers that help distinguish one task from another, such as a project name, account, session name, or repository. Use consistent labels where possible so that records from different sources can be compared without confusing unrelated work.

Decide which evidence is required. A session list can help establish which sessions are visible and active. Account and quota information can help with usage reviews. Cost information can help compare estimated usage across projects, while command histories can help identify repeated failures and their fixes.

Agree on the audit period, the questions to investigate, and how findings will be recorded. Distinguish information that is directly available from estimates or details that still need verification. This keeps conclusions tied to the evidence collected.

Finally, check the tools available on the machines being reviewed. Confirm their supported agent types, operating systems, account coverage, and available actions before relying on them for an audit. The goal is to collect enough relevant information to answer the audit questions without assuming every tool provides the same view.

Step by step

Define the audit scope. List the projects, accounts, agent types, and work periods to review. Decide whether the audit concerns session visibility, account usage, estimated costs, recurring failures, or a combination of these. Write down the questions each part of the review must answer.

Collect session records. Gather the available session information from the tools already used by the platform team. Look for details that distinguish projects and tasks. When sessions are spread across separate windows or accounts, identify a practical way to bring the available information together. Record any session types that are not represented in the collected view rather than assuming the list is complete.

Organize records by project and account. Use consistent project labels and account identifiers when reviewing the collected information. Separate sessions that belong to different projects, even if they use the same agent. Where a record cannot be confidently linked to a project, flag it for review instead of assigning it based on an assumption.

Review current activity. Check which sessions are visible and whether their status reflects the work being reviewed. Where a tool updates its list while sessions run, use that view to inspect current activity. For historical reviews, determine which past records are available and whether they cover the audit period. A live view and a historical record answer different questions.

Compare usage and cost information. Gather available account plans, quota information, and cost estimates. Separate usage figures from financial amounts, and note the basis of any cost estimate. If the audit needs actual billed amounts, verify them against the relevant billing records rather than treating an API list-price estimate as a final charge.

Investigate repeated failures. Review available command failures and the commands that resolved them. Look for recurring patterns that may justify changes to team guidance or agent rules. Check the surrounding task context before treating a repeated command as evidence of a platform-wide issue.

Record gaps and findings. Keep a record of the scope, sources reviewed, unresolved questions, and conclusions supported by the evidence. Identify missing session records, uncertain project assignments, and usage details that could not be verified. This makes it easier to distinguish a confirmed finding from a follow-up task.

Repeat the review when the scope changes. Revisit the process when projects, accounts, agent types, or review requirements change. Confirm that the selected tools still cover the relevant sessions and that their information supports the audit questions. A repeatable process helps platform engineers compare findings over time without assuming that every source has remained unchanged.

Tools that can do it

AgentHydra shows Claude Code, Codex and OpenCode sessions on your machine in one searchable list, updated live while they run. ^[1] It also shows cost at API list prices by day, model, project and account. ^[1] For account reviews, it lists each Claude and Codex account with its plan and 5-hour and weekly quota, and can open, focus, quit, create and delete Claude Desktop instances. ^[1] Its command-line tool prints the same cost numbers as JSON. ^[1] AgentHydra is free, with no account or subscription. ^[1] It is open source under the MIT License. ^[1] The tray app needs Windows; macOS and Linux builds exist but have not been verified. ^[1]

Existing agent workflows. The tools and views already available in an agent workflow are an option for collecting session information. Check whether they expose the sessions, project details, account information, and historical records required by the audit. Confirm the available coverage before treating the collected records as complete.

Session management tools. A separate session management tool may provide a way to organize session details. Check which agent types, session records, account details, and actions it supports before choosing it for cross-project reviews.

Usage and billing records. The usage information already available from relevant services can provide evidence for account and cost reviews. Check which time periods, projects, accounts, and cost definitions the records cover. Compare these details with session records before drawing conclusions about individual tasks.

Choose the approach according to the evidence needed. Session visibility, quota review, cost estimates, and historical investigation are different jobs. A tool that covers one area should not be assumed to answer every audit question.

Common mistakes to avoid

Avoid treating a visible session list as proof that every session has been captured. Check which agent types, accounts, machines, and time periods the chosen source covers. Record known gaps so that an incomplete view is not mistaken for a complete audit.

Avoid grouping sessions by project based on names alone. Similar task names can refer to separate projects or accounts. Use available project and account details, and flag uncertain assignments for further review.

Avoid treating estimated costs as actual charges. Cost figures based on API list prices are useful for comparison, but they should be distinguished from amounts recorded by the relevant billing service. Check the calculation basis and the period covered before reporting a financial conclusion. ^[1]

Avoid mixing current activity with historical evidence. A list that updates while sessions run can help inspect current work, but the audit should separately establish what past records are available and whether they cover the required period.

Avoid assuming that all agent tools support the same actions. Check which session types and account details each option provides. Confirm operating system support before deploying a tool across the platform team.

Finally, avoid recording conclusions without their evidence. Keep unresolved questions separate from confirmed findings, and document the sources used. This makes the review easier to repeat and helps other engineers understand what the collected information does and does not establish.

Things to check before you choose

Check whether the tool covers every agent type, account, and project included in the audit. Confirm whether its session view updates while work is running and whether historical information is available for the period under review.

Check how project and account details can be compared, whether cost figures are estimates or billing records, and which quota details are visible. If the audit includes recurring command failures, verify whether the chosen approach provides the evidence needed to examine those failures.

Confirm the supported operating systems, available session actions, and any required account access. For AgentHydra, check whether its session list, account information, cost estimates, and command-line output cover the specific evidence required for the review. ^[1] ^[1] ^[1] ^[1]

Frequently asked questions

How can platform engineers audit agent sessions across projects?

Start by defining the projects, accounts, agent types, and time period under review. Collect available session information, group records by project and account, and compare them with usage or cost evidence where needed. Record missing information and unresolved assignments separately. A repeatable audit process should state what sources were reviewed and which conclusions those sources support.

Can AgentHydra show agent sessions across projects?

AgentHydra shows Claude Code, Codex and OpenCode sessions on your machine in one searchable list, updated live while they run. ^[1] It also shows cost at API list prices by day, model, project and account. ^[1] Check whether the sessions and accounts visible on the machines under review cover the full audit scope before treating the list as complete.

Can AgentHydra help review agent usage by account?

AgentHydra lists each Claude and Codex account with its plan and 5-hour and weekly quota, and can open, focus, quit, create and delete Claude Desktop instances. ^[1] This information can support account and quota reviews. Check which accounts are relevant to the audit and whether the available details answer the usage questions being investigated.

Does AgentHydra show actual agent costs?

AgentHydra shows cost at API list prices by day, model, project and account. ^[1] Its command-line tool prints the same cost numbers as JSON. ^[1] These figures should be identified as API list-price estimates when reporting audit findings. Check the relevant billing records separately if the review requires actual billed amounts.

What should a cross-project agent session audit record?

Record the scope, projects, accounts, agent types, review period, sources examined, and findings supported by those sources. Include the available session identifiers and relevant usage or cost details when they are needed to answer the audit questions. Note missing records, uncertain project assignments, and estimates that need verification. Keeping unresolved questions separate from confirmed findings makes the review easier to repeat.

Free, with no account or subscription. ^[1]

Try AgentHydra

Sources

AgentHydra's own description of itself (the product of this site's publisher)