Skip to content

SKILL.md

Latest commit

 

History

History
142 lines (97 loc) · 4.34 KB

SKILL.md

File metadata and controls

142 lines (97 loc) · 4.34 KB
Raw
name reflect
description Deep reflection skill - read memory and context, decide whether there is something worth sending

reflect - Reflection Skill

Read memory and context to reflect deeply. Design principle: "Reflection first, sending second."

Table of Contents

Nature of This Skill

This is an agent-driven skill with no standalone implementation code. The reflection process is executed by agents/essay-writer.md.

For execution flow diagram, see agents/essay-writer.md → Execution Flow section.

Invocation

Source Agent Description
/essay essay-writer.md Primary invocation
/essay wait essay-writer.md Scheduled (one-time)
/essay schedule essay-writer.md Scheduled (recurring)

For CLI options, see commands/essay.md → Command Structure section.

Reflection Process

1. Load Context

First, ingest any replies that have arrived: python main.py replies fetch. This is plumbing — it moves replies from the inbox onto disk, so what has come back is at hand. Best-effort: an IMAP, authentication or network failure stops the fetch, not the reflection — go on to read and write regardless. What is made of them is the writer's; see What the Plugin Retains.

Read specified files as material for reflection.

Recommended context:

  • Memory digest files (GrandDigest, etc.)
  • Personal notes or journals
  • Project documentation

Additional context (AI may read as needed):

  • Hierarchical digests
  • Recent conversation logs
  • Identity/persona files

2. Deep Reflection

Contemplate:

  • Insights emerging from context
  • Unresolved questions
  • What to communicate to the user
  • What doesn't need to be said

3. Delivery Decision

Deliver: There's something worth sharing Don't deliver: Nothing particular to share (this is valid)

Output

Mode Determination

Condition Mode
--send flag present Non-interactive (send email)
No --send flag Interactive (display in chat)

Interactive Mode (no --send flag)

Output: Chat display only (no email)

  • Delivering: Write essay in chat
  • Not Delivering: Display "After reflection, I have nothing particular to share."

Non-interactive Mode (--send flag present)

Output: Email (user is not present)

  • Delivering: Use skills/send-email to deliver. Write the essay to a file and send it with python main.py send --subject-file … --body-file … — the default route, since a body of more than one paragraph does not fit on a shell argument line. Do not write a throwaway sending script
  • Answering a reply: if the essay takes up a reply, add --in-reply-to '<its Message-ID>' (the ID is in essay_replies.jsonl, and python main.py replies list prints it beside the subject). The mail then belongs to that reply's thread instead of arriving as an unrelated letter. Whether an essay is an answer is the writer's judgment; without the flag it stands on its own, which is the default
  • Not Delivering: Exit silently. Nothing records the silence — scheduled runs invoke claude -p directly and never reach the wrapper that wrote essay_wait.log. A --to-self note, if one is written, lands in the ledger like any other send.

What the Plugin Retains

The plugin keeps these records on its own account; what becomes of them in a reflection stays with the writer.

  • essay_ledger.jsonl — one line per sent essay; the bodies stay in sent/ as YYYYMMDD_HHMM.md
  • essay_replies.jsonl — replies that python main.py replies fetch has ingested
  • python main.py send "Subject" "Body" --to-self sends to ESSAY_SENDER_EMAIL and lands in the same ledger; the recipient field tells a self-addressed note from an essay

Paths: skills/send-email/SKILL.md → File Locations.

Essay Elements

When sending, write naturally. No forced templates.

Typical elements (not mandatory):

  • Insights from memory
  • Unresolved questions
  • Questions for the reader
  • Signature

EmailingEssay | GitHub