agentleFS
Sign inStart free
RandyRawgustRandyRawgust/ai-knowledge-base-template/.claude/skills/forge/SKILL.md

forge

Skill1 starsChanged 6 months agoMITView on GitHub

Mine session history (00 - Chats/, daily notes, brainstorms) for prompting patterns that recur 3+ times across separate sessions. Propose them as new skills, rules, or specialists. Use when Andy says forge, mine the sessions, what patterns am I repeating, find skill candidates, what should be a skill.

78 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use forge-ship, forge-review and 8 more.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • redteamlifeSkill

    forge-ship

    Use when the user asks to merge, release, promote, reconcile, or close out work in a repo containing `docs/forge/` or governed by FORGE. Not for per-task commits; those complete inside forge-execute-task.

    Also in 6 projects that use this

  • tonmoy007Skill

    forge-review

    Run a multi-dimensional parallel code review over the current change.

    Also in 8 projects that use this

  • weselbenSkill

    forge-docs

    Load before code changes, before writing docs or an ADR, when a user question exposes a missing how-to, and when closing a chunk of work; also loaded by forge after each squash-merge and by deep-research after a report.

    Also in 3 projects that use this

  • DailenGSkill

    forge-design

    Design, UX, and polish discipline for the forge lifecycle. Covers the design tier, the design brief in docs/DESIGN.md, UX requirements and their verification methods, the per-slice design pass, the UX defect register, and the release polish pass. Load whenever work touches a user-facing surface (GUI, terminal, library API, or operated service), whenever a UX observation needs disposing of, and whenever a release is in reach.

    Also in 3 projects that use this

  • DshamirSkill

    forge-plan

    Takes analysis.json + budget cap → derives the full phase sequence, hyperparameters, instance type, cost estimate, and acceptance thresholds. Emits plan.md (the single human gate document) and plan.json (the machine-readable spec that dispatch-v2 consumes). Refuses to plan if estimated cost exceeds budget×1.2.

    Also in 4 projects that use this

  • tonmoy007Skill

    forge-build

    Run Stage 6 of the Forge pipeline — implementation. Use when the user says

    Also in 3 projects that use this

The file

We split this skill into the rules it gives an agent

7 rules. 0 are common, 4 are distinctive, and 3 are very specific to this project.

  • 0

    Common

    Most of the 67 edited versions we read keep these.

  • 4

    Distinctive

    This author's own choices. Worth a look.

  • 3

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Write a forge report to 01 - Daily Notes/<YYYY-MM-DD> - Forge Report.md:

    line 50
  2. Very specific:

    Don't auto-create the skill/rule/specialist — Andy approves the promotion

    line 78
  3. Very specific:

    If approved, hand off to the relevant creator skill (/specialist-creator for specialists, direct edits to setup-claude-skills.ps1 for new …

    line 80
  4. Distinctive:

    Don't propose patterns that already exist (re-check the registry before reporting)

    line 83
  5. Distinctive:

    Don't propose anything that recurred only in one mega-session — that's narrowness, not pattern strength

    line 84
  6. Distinctive:

    Don't auto-promote — promotion is Andy's choice, you propose

    line 85
  7. Distinctive:

    Don't include the full source quotes — distill to the pattern, link to source

    line 86

“Common” is measured: kept by at least 6 in 10 of the 67 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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Forge

Cross-session pattern detector. Scans Andy's recent session history for patterns that have recurred enough to deserve promotion into a skill, rule, or specialist. Inspired by the FORGE archetype from the /goal video — the system-level version of the TCE evolution rule (add a symbol when it's been explained 3+ times).

When to use

Triggers: forge, mine the sessions, what patterns am I repeating, find skill candidates, what should be a skill, promote a pattern.

Best run after a substantive working week, or before a quarterly drift audit. The /retro skill captures lessons from ONE session; /forge looks ACROSS sessions.

Load

  • 03 - Skills & Rules/Rules/engineering-doctrine.md - research-first habits
  • 00 - Chats/ - session summaries (not full transcripts — those are too noisy)
  • 00 - Chats/brainstorms/ - past brainstorms
  • 00 - Chats/handoffs/ - past handoff notes
  • 01 - Daily Notes/ - last 30 days
  • 03 - Skills & Rules/Agents/Activity Log.md - last ~50 rows
  • Current skills list at .claude/skills/ (don't re-propose what exists)

Run

Phase 0 — Survey

Walk the source corpus. For each chat / brainstorm / daily note / activity row, extract: what task was being done, what instructions Andy gave repeatedly, what corrections he made, what manual steps recurred.

Phase 1 — Pattern detection

Cluster by similarity. A pattern qualifies as a candidate when:

  • It appears in ≥3 separate sessions (different days, different contexts)
  • It involves a workflow that took multiple turns to describe
  • Andy's correction pattern is similar across the instances
  • A skill could meaningfully wrap it (not just trivial 1-line ops)

Phase 2 — Propose

Group candidates by promotion target:

PromotionWhen
New skillThe pattern is invocable via a trigger phrase + has a clear inputs→outputs shape
New ruleThe pattern is a recurring habit / convention (not invocation-shaped)
New specialistThe pattern has enough scope + judgment to warrant its own agent
Extension to existing skill/specialistThe pattern is a feature gap in something we already have
No promotionPattern recurs but doesn't earn a slot (under 3 hits, or pattern is too narrow)

Output

Write a forge report to 01 - Daily Notes/<YYYY-MM-DD> - Forge Report.md:

---
type: forge-report
created: <MM-DD-YY>
sessions_scanned: <count>
candidates_found: <count>
candidates_proposed: <count>
tags: [forge, doctrine-evolution]
---

# Forge Report -- <MM-DD-YY>

## Candidates proposed
| Pattern | Recurrence | Promotion target | Notes |
|---|---|---|---|
| <one-line description> | <N sessions> | new skill /<name> | <why> |

## Candidates rejected (under threshold)
- <pattern>: only <N> hits, watch for more

## Recommended next action
- <which candidate to build first, why>

After

  1. Surface the top 1-3 candidates with concrete proposals for what they'd look like
  2. Don't auto-create the skill/rule/specialist — Andy approves the promotion
  3. Append Activity Log row
  4. If approved, hand off to the relevant creator skill (/specialist-creator for specialists, direct edits to setup-claude-skills.ps1 for new skills, direct edits to Rules/ for new rules)

Don't

  • Don't propose patterns that already exist (re-check the registry before reporting)
  • Don't propose anything that recurred only in one mega-session — that's narrowness, not pattern strength
  • Don't auto-promote — promotion is Andy's choice, you propose
  • Don't include the full source quotes — distill to the pattern, link to source

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.

Check your setup, free