Index of every page on this site: llms.txt
All skills
Run it yourself

Finds repeated manual work in activity records and proposes automations with source evidence. Starts from a team Slack channel, a mailbox, an issue tracker, or supplied work records. The owner reviews each proposal. Use when the user wants automation opportunities found in their own or their team's actual activity: "mine this channel", "what could we automate", "find repeated manual work in these records". Do not use when the user already knows what to build, or wants ideas without granting read access; that is an interview, answer it directly.

Use this Skill: https://skilld.dev/gh/n8n-io/automation-mining-skill/automation-mining

Nothing lands on disk. Nothing to clean up.

Fork this Skill

Edit a local copy. It keeps the author and licence.

referencestrails.md

≈1.4k tokens on demand. Your agent reads this file only when SKILL.md points to it.

Trails — what earns a new surface, and how it is read once earned

A trail is a specific pointer from the surface you read to one you haven't. It earns an ask, never an automatic open (SKILL.md, step 5). The same ask applies when the host cannot read that surface: name the connection needed and both payoffs; the user decides. A closed surface never cancels a trail — coverage lists it as unconnected, asked. Read only within the scope the owner approved. For verification only, use Correlate and stop. When the owner also approves mining, use Correlate, then Mine:

  • Correlate — bounded probes that support or kill the patterns you already hold. Cheap, and it tells the owner whether the first pass was right.
  • Mine — the same complete read the first surface got, in this surface's own unit and approved time window. Follow the listing and reading rules in SKILL.md step 3; do not sample only for known patterns. This is where new patterns come from; the first surface's ceiling is the coordination work it shows.

Then one merged shape check. Name new trail types when you meet them; every unfollowed trail goes in the coverage section.

Automation footers and bot posts

Recurring posts signed by a workflow or bot, scheduled digests, "posted by".

  • Points to: the automation estate — the workflow or job definitions, their run history, the replies under their posts.
  • Correlate: which of your patterns the automation already covers, and how well — count the corrections and complaints under its output.
  • Mine: unit = a workflow or scheduled job. Manual fixes after runs, re-posts, hand corrections of its data, jobs nobody reads.
  • Ask: "3 recurring posts here come from workflow X and 6 replies call its numbers wrong — reading the workflow and its runs tells me whether the fault is the workflow or its data, and what else runs there."

Artifact gravity

Three or more units orbit one document, board, sheet, or page.

  • Points to: that artifact and its neighbours.
  • Correlate: is it fresh, owned, and does it hold what people asked for. A discovery failure (the answer existed, nobody found it) is automatable; an absence (no document exists) is a gap for the report, not a proposal.
  • Mine: unit = a page, row, or record. What is edited by hand on a cadence, pasted in from elsewhere, or asked about again.
  • Ask: "Half the doc questions end at that Notion space — checking it tells me whether the docs are missing or just unfound, and what the space itself repeats."

Relay chain

One object carried by hand across two or more systems: thread into ticket, doc into sheet, email into CRM.

  • Points to: the second system.
  • Correlate: count the hops and who carries them; does the object arrive intact; how long does it wait between systems.
  • Mine: unit = a ticket or record. What repeats inside that system on its own, independent of the first surface.
  • Ask: "11 threads were re-typed into the tracker by hand — reading the tracker tells me which fields travel, and what the tracker's own routine looks like."

Dispute on automated output

"The numbers look wrong", "the bot got it wrong", a human redoing what a machine posted.

  • Points to: the automation and its data source.
  • Correlate: how often the output is corrected by hand; is the failure in the automation or in its input.
  • Mine: unit = a run or report. Which corrections repeat.
  • Ask: "6 of 14 leaderboard posts drew a correction — reading the workflow and the source table tells me whether to fix the query or the data, and what else that workflow's runs get corrected by hand."

Lag between a change and later questions

Something ships; the questions about it arrive days or weeks later.

  • Points to: the change stream — merged changes, release notes, announcements.
  • Correlate: match the questions to the changes that caused them.
  • Mine: unit = a change. What has to be told to whom after each one — the repeated hand-off or announcement.
  • Ask: "A UX change merges and this channel asks 'bug or by design?' one to two weeks later — reading merged changes tells me which ones will raise questions before they do, and what has to be told to whom after every merge."

Your own record with this user

The surface points back at you — a ritual they ask you to run, a channel where your output lands, a report you wrote — or step 1 offered the record and they took it.

  • Points to: the host's session store or chat history, the memory you keep for them, your past reports and the verdicts on them.
  • Correlate: does the pattern you hold on their channel also arrive as an ask to you — the same lookup, relay, or prep requested of the agent by hand? Count those asks and date them against the channel's units.
  • Mine: unit = a session, read through a condensed view the merger generates (typed asks, tools called, files written); the raw log opens only to verify. What they ask again, what they paste in, what you run for them on a cadence, what the host runs on a schedule and how often it fails. "Resolved by" names what actually ran, not what was called.
  • Ask: "Fifteen pre-call sheets in 26 days arrived here as typed asks — reading my own sessions tells me what else you hand me on a cadence, and the calendar tells me which calls never got one."

Source: SKILL.md on GitHub

No rule matched.

skilld matched fixed text patterns in SKILL.md and file names. Patterns miss obfuscated code.

skilld run checks every file with the same patterns. It asks for approval before it loads a Skill with a behavior marked Needs approval.

1 warning1mo3 checks · Risk SAFE
  • Gen Agent Trust Hub1mo

    The skill is a legitimate automation mining tool that follows high security standards. It includes explicit rules to prevent prompt injection, protect sensitive data, and maintain a read-only posture on external services.

  • Socket1mo

    No alerts

  • Snyk1mo

    Risk: MEDIUM · 1 issue

Signed by skilld at c612681. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub last month.

Activeupdated 2 months ago
metadata
{
  "version": "1.0.0"
}

README badge

README badge for n8n-io/automation-mining-skill