Developer

Skill pablodiazjorge/prompt-forge/packages/custom/.github/skills/developer

Drop-in toolkit that gives AI coding agents persistent memory across sessions. Agent Skills, cross-session learning loop, and knowledge persistence. Zero dependencies, no backend.

Install

npx -y skills add pablodiazjorge/prompt-forge --skill developer

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Continuous issue tracking at the end of every coding iteration. Scans the chat for errors, API discoveries, workarounds, and patterns. Cross-references with the issue registry and creates or updates issues in knowledge/issues/open/. Safe and non-destructive — NEVER promotes issues, NEVER modifies skills or memory files. Runs by default at session end or when the user says "analyze this session", "what did we learn", or "save lessons".

The file declares its own license as MIT. That is the author’s claim about this one file, and it is not the same thing as the license GitHub reports for the repository, which is listed with the other numbers below.

SKILL.md

5.4 KB, as published. Nobody here has run it

Developer: Continuous Issue Tracking

Trigger

Run at the END of every significant coding iteration OR when the user explicitly requests it. Do NOT run mid-task — wait for a natural stopping point.

This skill is loaded by default at session start. It is safe and non-destructive: it only writes to knowledge/issues/open/ and knowledge/issues/INDEX.md.

Algorithm

Phase 1 — Scan Current Chat

Review the current conversation for these signal types:

Signal	Examples	Priority
Error fixed	"PowerShell interpretó `{{ }}` mal", "ngx-translate v18 no tiene `forRoot()`"	HIGH
API discovery	"`linkedSignal` requires Node ≥22", "`provideZonelessChangeDetection()` is stable in Angular 22"	HIGH
Workaround found	"Usa `create_file` en vez de `Set-Content` para templates"	MEDIUM
Pattern repeated	Same error occurred 2+ times in this session alone	HIGH
Project quirk	"Este monorepo usa npm workspaces", "El `.gitignore` excluye `.github/`"	MEDIUM
Tool limitation	"Copilot no expone `usage.prompt_tokens` directamente"	LOW
User preference	"Prefiere comandos separados con `;` no `&&`"	MEDIUM

Phase 2 — Cross-Reference with Issue Registry

Read knowledge/issues/INDEX.md — get all existing issue IDs, titles, categories
For each signal from Phase 1, attempt to match against existing issues:
- Match by keyword overlap (title + description)
- Match by category + similar error pattern
- Use grep_search on knowledge/issues/open/ for related content

Phase 3 — Create or Update Issues

NEW issue (no match found)

Generate ID: ISSUE-YYYYMMDD-XXXX where XXXX = first 4 hex chars of a UUID
  PowerShell: [guid]::NewGuid().ToString().Substring(0, 4)
  Example: ISSUE-20260624-a1b2
Create knowledge/issues/open/ISSUE-YYYYMMDD-XXXX.md using TEMPLATE.md
Set occurrences=1, certainty=low, status=open
Set created_by to "developer" (or current agent identifier)
Insert a new row into INDEX.md → Open Issues table
If you have session token data, optionally add it to Evidence section

IDs are UUID-based — no need to read INDEX.md first. Zero collision risk across multiple developers working simultaneously.

UPDATE existing issue (match found)

Increment occurrences by 1
Update last_seen to today's date
Append current session info to sessions array
Re-evaluate certainty:
  - 1 occurrence → low
  - 2 occurrences → medium
  - 3+ occurrences → high
Update the issue file in open/ using replace_string_in_file
Update INDEX.md counts and row data

INDEX.md merge conflicts

If two developers add rows to INDEX.md simultaneously, a Git merge conflict will occur on the table body. Resolution: keep ALL rows from both branches; re-count directories to update Summary numbers.

Report

Output a summary table:

## Developer Report — YYYY-MM-DD

| Action | ID | Title | Category | Detail |
|--------|----|-------|----------|--------|
| 🆕 New | ISSUE-001 | ... | ... | Created in open/ |
| 📈 Updated | ISSUE-002 | ... | ... | 1→2 occurrences, certainty low→medium |

Important Rules

NEVER promote issues. Promotion is the responsibility of the ai-engineer skill. Your only output is issue files in open/ and updates to INDEX.md.
NEVER modify skills or memory files. You are a read-only observer of .github/skills/ and /memories/. Do not touch them.
NEVER create a skill for something the model already knows — if it's common knowledge (e.g., "use const not var"), skip it entirely.
Keep issue files concise — the TEMPLATE.md format is the maximum, not the minimum. Don't write essays.
Session isolation — the sessions array in frontmatter stores session UUIDs. Use session_store_sql to find the current session UUID if available, otherwise use a date-based identifier.
Always populate created_by — set it to the agent name or session identifier so the ai-engineer can trace issue origins during review.

Handoff to AI Engineer

When an issue reaches certainty: high (3+ occurrences), the ai-engineer skill handles promotion. The developer skill does NOT decide what to promote or where — it only records the signal strength. Run the ai-engineer skill as a subagent when the user asks to organize knowledge, clean up issues, or promote patterns.

File Paths Reference

Resource	Path
Issue registry index	`knowledge/issues/INDEX.md`
Issue template	`knowledge/issues/TEMPLATE.md`
Open issues	`knowledge/issues/open/ISSUE-XXX.md`
Promoted issues	`knowledge/issues/promoted/ISSUE-XXX.md`
Discarded issues	`knowledge/issues/discarded/ISSUE-XXX.md`