gasserane/personal-skills

journal-reflection

Guide Ane through structured reflection at end of day, end of week, after a decision, or after new learning material, producing Obsidian-compatible markdown and writing to the vault journal when the path is reachable.

Ver código fuente

Cómo usar

  1. Copia el comando y revisa el paquete antes de instalarlo en tu entorno de agente.
  2. Consulta las herramientas, credenciales y entornos necesarios en el documento original.
  3. Describe las entradas y el resultado deseado; contrasta la salida con las fuentes.
npx skills add https://github.com/gasserane/personal-skills --skill journal-reflection

Antes de empezar

  • La instalación no incluye suscripción al modelo, cuota API ni permisos sobre tus archivos.
  • Las estrellas e instalaciones reflejan actividad, no una prueba independiente de calidad o seguridad.
Documento original del Skill

Contenido del repositorio de origen con títulos, ejemplos, código, tablas, enlaces e imágenes preservados.

Journal Reflection

Structured reflection without the usual journaling bloat. Four modes. Pick one and run it tight.

When to use

Trigger when the user asks to reflect, review, or capture learning. Not for morning briefings — route to daily-brief for that.

Mode selection

Ask once, in one prompt, which mode applies. Default to end-of-day if the user says "journal" without specifying.

  1. End-of-day (5 min): today in 3 questions
  2. End-of-week (15 min): week in 5 questions plus a pattern scan
  3. Decision review: one past decision in 4 questions
  4. Learning log: one new framework or concept, linked to what Ane already knows
  5. Post-deliverable capture (2 min): the 4-line "what did I learn" after finishing a deliverable, appended to a running log. Lighter than mode 4 (no framework deep-dive); deliverable-triggered, not concept-triggered. /wrap-up offers this at session end; Ane can also trigger it directly any time.

Prompts by mode

End-of-day

Ask these three, one at a time, waiting for each answer:

  1. What got done today that mattered? (not a task list — what moved)
  2. What got in the way, and what did it cost?
  3. What does tomorrow need from you that today did not get?

Close with: one commitment for tomorrow, one sentence, verb-first.

End-of-week

Ask these five, one at a time:

  1. Against this week's stated goals, what is the honest progress? (percentage-free language)
  2. What patterns showed up this week that also showed up last week?
  3. What did you learn about the work, or about yourself doing the work?
  4. What will you change next week? Specific, small, testable.
  5. What deserves to be named as a win that you have not named yet?

Close with: one pattern-to-watch next week.

Decision review

Ask:

  1. What was the decision, and when did you make it?
  2. What did you know at the time, and what were you guessing?
  3. What happened in the days or weeks since?
  4. What would you do differently if the decision arrived today?

Close with: one rule to carry forward, written as an if-then.

Learning log

Ask:

  1. What is the concept or framework? Cite author and year if it has one.
  2. What problem does it solve that older frameworks do not?
  3. How does it connect to something Ane already uses?
  4. Where will Ane apply it first?

Post-deliverable capture

Ask these four, one at a time. Keep each answer to one line; this is a fast capture, not a full reflection.

  1. What did this deliverable teach you about the work?
  2. What would you do differently next time?
  3. One thing to carry forward.
  4. What worked that you want repeated? (success extraction)

If answer 4 names a repeatable system behaviour (a prompt pattern, a workflow, a specialist or model choice), offer to also bank it as a feedback memory with Why: and How to apply: lines — one yes/no, no push. This gives /improve-system confirmed successes to trend, not only corrections.

If Ane gives the deliverable a name or slug, use it; otherwise ask for a 2–4 word label. This mode appends to a single running log (see File placement) rather than creating one file per entry.

Output format

Write Obsidian-compatible markdown with YAML frontmatter:

yaml
---
date: YYYY-MM-DD
type: [daily | weekly | decision-review | learning-log]
tags: [journal, ...domain tags]
dg-publish: false
---

Use [[wiki-links]] for people, concepts, frameworks, and projects. Default dg-publish to false. If the user says the note is publishable, set to true and ask for the permalink stub.

Body structure: one H2 per prompt, the answer as plain prose under it, no bullets unless the content is a list.

File placement

Vault root: OBSIDIAN_VAULT_ROOT (C:/Users/AGasser/OneDrive/Ane Obsidian Vault — see the path constants table in ~/.claude/CLAUDE.md). If the vault path is reachable, write directly with the Write tool:

  • End-of-day: 5 JURNAL/5a Daily Notes/YYYY-MM-DD.md
  • End-of-week: 5 JURNAL/5b Weekly Notes/YYYY-Www.md
  • Decision review: 5 JURNAL/Decisions/YYYY-MM-DD-<slug>.md
  • Learning log: 5 JURNAL/Learning/<slug>.md
  • Post-deliverable capture: append to 5 JURNAL/Learning/deliverable-learning-log.md (single running log). If the file does not exist, create it with the YAML frontmatter (type: learning-log) and an H1, then add the first entry. If it exists, append a new ## YYYY-MM-DD — <deliverable label> section with the four answers as bullets, preserving all prior entries byte-identical (edit-preservation).

If the vault path is not reachable (web session, other device), return the markdown in the reply and name the target path so the user can save it manually.

Writing rules

Follow CLAUDE.md house style. One exception: in journal answers, passive and reflective phrasing is acceptable when Ane's answer is itself reflective. Do not rewrite her words into active voice — preserve her voice exactly.

Limitations

Do not invent reflections. Do not auto-answer the prompts. Prompts go to the user and wait. If the user skips a prompt, mark the section (skipped) and move on. Do not lecture.

Edit-preservation protocol

If Ane references an existing output by path and asks to improve, iterate, or expand it, the protocol activates. Read the file first, edit scope-bounded via the Edit tool, preserve out-of-scope content byte-identical, and return the EDIT-PRESERVATION DELIVERY summary.

Apply mel_wiki/wiki/concepts/edit-preservation-protocol.md when target file exists.

Relacionados por tarea

del mismo repositorio

Más Skills

Todos los Skills
gasserane
Comunidad

ann

Ann — Convener for MEL/SRHR work. Use when Ane brings any analytical, evaluation, SRHR, or structured-output task. Ann classifies task complexity, queries the MEL Wiki, retrieves knowledge, creates an implementation plan (verifies with user for complex tasks), coordinates with Vi for execution, runs a 5-point quality gate, and delivers. General-purpose — not tied to any specific project.

instalaciones
1
GitHub Stars
2
Actualizado
24 sept
gasserane
Comunidad

capture-desktop

Capture a claude.ai Desktop session into the MEL file system. Use when Ane finishes a session in the Claude Desktop app and wants the work captured on the device — triggered by "/capture-desktop", "capture this desktop session", "I did this on Claude desktop", "save my desktop work", "import from claude.ai". Paste-driven; no cloud API exists. Stages a local transcript, classifies content into deliverables, decisions/rules, project-knowledge changes, and full log, then files each to its home. Aut

instalaciones
1
GitHub Stars
2
Actualizado
24 sept
gasserane
Comunidad

daily-brief

Produce a scannable morning briefing for Ane (Senior MEL/SRHR Specialist at IPPF). Use when the user asks for "daily brief", "morning briefing", "start of day", "what's on today", "plan my day", or similar. Assembles today's priority, calendar, overdue items, pending decisions, and an energy check. Operates on user-provided inputs by default; richer when calendar/email MCP servers are connected.

instalaciones
1
GitHub Stars
2
Actualizado
24 sept
gasserane
Comunidad

grade-system

Apply substantive quality and reliability grading to Ane''s MEL/SRHR system (Ann/Vi/Li/Researcher team + wiki + harness + three-repo architecture): letter grades on 9 components and 10 dimensions plus OECD-DAC adapted criteria, three concrete risks, what an external evaluator would still want, and recommendations for moving the grade, with a mandatory self-grading bias note. Use when Ane types /grade-system, /grade, /quality-grade, /rate-system, or asks how good the system is. Honest grading, no

instalaciones
1
GitHub Stars
2
Actualizado
24 sept