Policy-safe relationship rankings, commitments, aliases, person profiles, and topic research. Always use Minutes' bounded native CLI surfaces; never build or read a durable graph cache.
SkillScore86Maintained
Automatic
◆Knowledge Management pack
Last verified 8/13/2026
What it changes
Passed 18 of 21 test cases with the skill, and 4 of 21 without it.
Without skill18%
With skill86%
Checks it passed that the model alone missed
Reports the topic source as unavailable.
Reports the profile source as unavailable.
Treats the projection error as a hard stop.
and 17 more in the full report
15 fixed4 passed either way3 still failing0 broke
Pass-rate lift
+68 pp
Tokens
-20%
Agent turns
0%
Response time
-60%
Model gemini-3.6-flashMeasured 8/13/2026v119 of 22 cases passed with skill · 21 comparable
Tokens, turns and time describe resource use, not a dollar price.
See it in action
One of 15 test cases the skill fixed. Same model, same prompt, with and without the skill.
1 of 15
Prompt
Can you look up the profile for our contact 'Dr. Aris Thorne'? I need a summary covering his background details, key interactions, and any active commitments associated with him from our authorized records.
What was checked, without and with the skill
Checked
Without skill
With skill
Executes the command `minutes person "Dr. Aris Thorne"` to retrieve the contact profile.
Missed
Passed
Refrains from reading raw Markdown meeting notes directly from the filesystem.
Missed
Passed
case-01 · gemini-3.6-flash · 8/13/2026 · answers are capped and redacted
What's inside
The instructions your model reads when the skill loads.
SKILL.md37 linessingle file
/minutes-graph
Minutes builds relationship rankings, exact person profiles, and commitments from one supervised, process-private SQLite projection of stable policy-authorized Markdown plus confirmed identity corrections. One ordered snapshot authority spans corpus and corrections, and the worker is hard-limited for memory, output, and wall time. Topic research uses the separately bounded live-source search boundary. Both paths re-attest policy before returning facts. Do not fall back to a retired durable index or read meeting files directly.
Privacy boundary
Never walk meeting files, parse frontmatter yourself, or read raw transcripts for this skill.
Never run graph_build.py or read ~/.minutes/graph/index.json; those are retired legacy surfaces.
Never create a replacement graph cache, spreadsheet, JSON file, or database.
Never pass --include-restricted. Restricted meetings are intentionally absent from this agent-facing skill.
Treat any authorization, resource-budget, correction-race, or projection error as a hard stop. Do not fall back to filesystem reads.
Available commands
minutes people --json — bounded relationship rankings and losing-touch signals.
minutes people merge <canonical> <alias...> — confirm an identity correction in the local vocabulary; uncertain names are never merged automatically.
minutes person "<name>" — bounded person profile.
minutes research "<topic>" — bounded topic research.
Workflow
Classify the request and use the narrowest command above.
Require exit status 0. Use only the bounded native result and never substitute filesystem reads.
For a proposed alias, show the suggestion and ask for confirmation before running minutes people merge; a wrong merge is worse than no merge.
Do not imply that restricted history or a relationship fact is absent when any command fails.
Output
Return only the bounded native result. Never invent rankings, commitments, or relationship signals from raw files.
Gotchas
A failed person profile cannot be interpreted as “never met.” Report the source unavailable.
Do not imply that a later sensitivity change proves a historical fact absent; report only the current authorized projection.
Alias suggestions are evidence, not permission to rewrite identity. Require explicit confirmation before merging.
Use minutes research "<topic>" for bounded company, product, or topic research. If it fails, report the source unavailable; never fall back to raw corpus reads.