audit-memory
Skill59 starsChanged 12 days agoMITView on GitHub
Check agent instructions, memory files, and project docs for old information, incorrect symbols, and context size. Use only when the user explicitly requests an audit of these files.
4 other projects use a version of this.
Adopted by
Most recently taken up by
- gspencergoog/skillstheir own version, first seen Nov 16, 2026
- nicholasjconn/skillstheir own version, first seen Nov 15, 2026
- gethrbr/deckhandtheir own version, first seen Nov 15, 2026
- PRATHAMESHG1234/claude-skill-audit-memorytheir own version, first seen Nov 8, 2026
Other versions4 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 4
- gethrbr/deckhand★ 3same name
- gspencergoog/skills★ 3same name
- nicholasjconn/skills★ 1same name
- PRATHAMESHG1234/claude-skill-audit-memory★ 1same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository17 other files it gives its agents.
- AGENTS.mdAGENTS.md
- clawkerd/AGENTS.mdAGENTS.md
- controlplane/AGENTS.mdAGENTS.md
- docs/AGENTS.mdAGENTS.md
- test/AGENTS.mdAGENTS.md
- .github/copilot-instructions.mdCopilot instructions
- agent-filesSkill
- cli-outputSkill
- dev-checksSkill
- firewall-uatSkill
- initiativeSkill
- monitoring-checksSkill
- store-editorSkill
- test-hunterSkill
- wizard-commandsSkill
- writing-testsSkill
- code-reviewSkill
The file
We split this skill into the rules it gives an agent
12 rules. 0 are common, 5 are distinctive, and 7 are very specific to this project.
0
Common
Most of the 4 edited versions we read keep these.
5
Distinctive
This author's own choices. Worth a look.
7
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 9
Use this procedure only when the user explicitly requests an audit of agent instructions, memory, or project documentation.
- Distinctive: line 13
Check the shared agent files and the active tool’s memory files.
- Very specific: line 32
Check every CLAUDE.md link: its relative target must be the sibling AGENTS.md.
- Very specific: line 58
Run serena memories check when the CLI is available.
- Very specific: line 67
Broken links: Check that every (filename.md) reference in MEMORY.md points to an existing file.
- Very specific: line 68
Stale project memories: Read each project_*.md and firewall_*.md file.
- Distinctive: line 70
Context limits: Check the active tool’s documented load limit.
- Very specific: line 74
Check if docs/docs.json navigation groups reference files that actually exist in docs/ and docs/cli-reference/
- Very specific: line 80
Read .serena/memories/memory_maintenance.md and follow its memory graph rules.
- Distinctive: line 86
Check for contradictions between always-loaded context files:
- Distinctive: line 96
Check for mentions of outdated components, patterns, or practices
- Very specific: line 119
SCOPE: Always-loaded rules that should have paths: frontmatter
“Common” is measured: kept by at least 6 in 10 of the 4 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Audit Memory
Use this procedure only when the user explicitly requests an audit of agent instructions, memory, or project documentation. Do not start an audit merely because another task reads or changes those files.
Check the shared agent files and the active tool’s memory files. See REFERENCE.md for package and memory checks.
Steps
1. Inventory
List the source files below. Do not count a symbolic link as a second copy:
AGENTS.md(root)cmd/**/AGENTS.md,internal/**/AGENTS.md,test/**/AGENTS.md, andpkg/**/AGENTS.mdcontrolplane/**/AGENTS.md,clawkerd/AGENTS.md.agents/skills/*/SKILL.mdand their sidecar files.serena/memories/**/*.mddocs/docs.json,docs/custom.css,docs/favicon.svg— Mintlify site config, theme, and favicondocs/*.mdx— Hand-authored Mintlify pagesdocs/cli-reference/*.md— Auto-generated CLI reference (never edit directly; generated via Makefile, freshness checked in CI)
For each file, report: path, line count (wc -l), estimated tokens (wc -c / 4).
Check every CLAUDE.md link: its relative target must be the sibling AGENTS.md. Check the links inside .claude/ and .codex/ against .agents/skills/agent-files/SKILL.md.
Group files by the active tool's documented loading behavior:
- Initial instructions: root
AGENTS.mdthrough the tool's native entry path - Package instructions:
cmd/**/AGENTS.md,internal/**/AGENTS.md,test/**/AGENTS.md,pkg/**/AGENTS.md - On-demand:
.agents/skills/*/SKILL.md(loaded on description match) and.serena/memories/**/*.md(loaded throughmem:references) - WIP tracking:
.serena/memories/**/*.md - Mintlify site:
docs/docs.json,docs/custom.css,docs/*.mdx,docs/cli-reference/*.md
2. Freshness Check
For each AGENTS.md, compare its last commit date with the last commit date of Go files in its directory (git log -1 --format=%cs -- <path>). List files whose source changed after the instruction file.
3. Symbol Accuracy
For each package AGENTS.md found in the inventory:
- Read the file
- Extract all backtick-wrapped Go identifiers (pattern: single backtick-wrapped words matching
[A-Z][A-Za-z0-9]*— exported symbols) - For each identifier, grep for it in
*.gofiles in the same directory - Report:
- Missing: identifiers documented but not found in Go source (renamed or deleted)
- Undocumented: exported Go symbols (
^func [A-Z],^type [A-Z],^var [A-Z],^const [A-Z]) in the directory not mentioned in the AGENTS.md. ExcludeTest*andBenchmark*functions — these don't belong in AGENTS.md.
4. Serena Graph Validation
- Run
serena memories checkwhen the CLI is available; otherwise resolve everymem:reference by hand. - Confirm
corelinks each domain memory and each referring line says what the target covers. - Flag behavioral or situational content in a memory; it belongs in a skill.
5. Auto Memory Audit
Check Serena's .serena/memories/ graph and the active tool's memory directory when it is available. Skills and package AGENTS.md files are not auto memory. Claude Code uses ~/.claude/projects/*/memory/. Do not assume that Codex uses that structure. If a memory source is unavailable, report that limit:
- MEMORY.md index completeness: List all
.mdfiles in the directory. Flag any not referenced inMEMORY.md(unindexed). - Broken links: Check that every
(filename.md)reference inMEMORY.mdpoints to an existing file. - Stale project memories: Read each
project_*.mdandfirewall_*.mdfile. Flag those whose descriptions no longer match reality (e.g., "ready for planning" when work is complete). - Frozen dates: Flag any
currentDateor hardcoded date blocks inMEMORY.md— these rot. - Context limits: Check the active tool’s documented load limit. Claude Code loads the first 200 lines of its auto-memory index; do not apply that limit to other memory formats.
6. Mintlify Docs Consistency
- Check if
docs/docs.jsonnavigation groups reference files that actually exist indocs/anddocs/cli-reference/ - Flag hand-authored pages (
docs/*.mdx) that reference outdated commands or config keys (spot-check against.clawker.yamlschema and CLI command tree) - Skip
docs/cli-reference/*.mdfor content accuracy — they are auto-generated
7. Serena Memory Staleness
Read .serena/memories/memory_maintenance.md and follow its memory graph rules. Check .serena/memories/**/*.md for broken mem: references, obsolete current guidance, and unsupported claims.
Retained history and research can describe completed work. Do not mark a file for deletion only because it contains a completion marker or is more than 30 days old. Check its purpose and links first.
8. Contradiction Detection
Check for contradictions between always-loaded context files:
- Root AGENTS.md vs Serena
conventions: Identify instructions that appear in both. Flag exact duplicates (wasted context) and conflicting statements. - Root AGENTS.md vs global instructions (for example,
$CODEX_HOME/AGENTS.mdor~/.claude/CLAUDE.md, when present): Check for conflicting behavioral directives (e.g., "pivot on tech debt" vs "surgical changes only"). - Within root AGENTS.md: Flag repeated information (e.g., same fact stated twice in different sections).
9. Architecture and Design Accuracy
- Identify changes in architecture, design, CLI commands, test harnesses, or test doubles from the freshness check, git status, or commit messages
- For each reference skill (
writing-tests,cli-output,dev-checks,agent-files) and each design memory (architecture,design,key-concepts,repo-structure,project-guide):- Check for mentions of outdated components, patterns, or practices
- Flag files that likely need updates based on the nature of the changes
10. Context Budget
Report totals against budgets:
| Category | Budget | Actual |
| Root AGENTS.md + always-loaded rules (no paths:) | < 500 lines | ? |
| Total always-loaded (root + rules + global) | < 800 lines | ? |
| Each individual AGENTS.md | < 200 lines | ? |
Flag any files exceeding their budget.
11. Recommendations
Output a prioritized action list using these categories:
- DELETE: Completed WIP memories, stale auto-memory files
- UPDATE: Stale docs (from freshness check) or docs with missing/wrong symbols
- FIX: Contradictions between always-loaded files
- TRIM: Files exceeding context budget
- SCOPE: Always-loaded rules that should have
paths:frontmatter - ADD: Packages in
internal/orpkg/with significant Go files but noAGENTS.md
Format each recommendation as:
[ACTION] path/to/file — reasonSort by priority: DELETE > FIX > UPDATE > TRIM > SCOPE > ADD.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

