Skip to content

Prompts Reference

This page documents server-side prompts exposed via the MCP Prompts API. These prompts do not call LLMs; they return message sequences your client can feed to an LLM.

worklog_entry

  • Purpose: Generate an engineering worklog entry for a given day or period. If tool-calling is available, it orchestrates Git and GitHub data gathering before synthesis and then persists the result into your worklog file.
  • Arguments:
  • date (string): Target day or period. Accepts YYYY-MM-DD, a range (YYYY-MM-DD..YYYY-MM-DD), or relative expressions like "yesterday", "last 2 days".
  • inputs (string, optional): Free-text notes to incorporate.
  • Behavior highlights:
  • Gathers commits per repository using git MCP tools (get_commits_by_date, get_enriched_commits, get_remote_origin, determine_commit_url_prefix).
  • Correlates commits with conversations (get_recent_conversations) and with pull requests when PR numbers are found.
  • Produces structured sections: Goals & Context, Technical Work, Metrics, Key Decisions, Impact, Open Items, Learnings.
  • Persists one block per date by calling append_to_markdown with update_mode=true so entries merge idempotently.
  • Optionally writes a Weekly Summary for the Monday..Sunday window when a full week has completed and there are newer entries past the Sunday.
  • Returns: list of messages like [{"role": "system"|"user", "content": "..."}] suitable for your LLM client.
  • See code: seev.prompts.worklog_entry_prompt.

Slash command usage

  • Syntax: /worklog_entry <date> [inputs]
  • Positional arguments: date is required; inputs is optional.
  • Quoting: wrap strings that contain spaces in quotes. Double or single quotes are both allowed.
  • Spacing: use a single space between tokens.

Examples (today is 2025-10-26): - /worklog_entry 2025-10-25 "Wrapped up feature X; reviewed PR #123" - /worklog_entry "last 2 days" "Pair-programmed on the parser; validated caching behavior" - /worklog_entry 2025-10-20..2025-10-26

Expected tool calls (handled by your assistant): - get_tracked_repositories_config - get_commits_by_date (per repo) - get_enriched_commits (optional) - get_recent_conversations - get_remote_origin → determine_commit_url_prefix - append_to_markdown (once per date)

conversation_summary

  • Purpose: Create a concise summary of the current task/conversation and persist it by calling the record_conversation_summary tool.
  • Arguments:
  • date (string, optional): ISO calendar date (YYYY-MM-DD). If omitted/blank, the server resolves it to today's local date when rendering the prompt.
  • title (string, optional): Title used when creating a new conversation. If not provided and conversation_id is null, the LLM should derive a short title from the first sentence (<= 80 chars, no newlines).
  • conversation_id (integer, optional): Existing conversation id to append the summary to; when provided, title is ignored.
  • inputs (string, optional): Notes, highlights, or raw text to base the summary on.
  • Behavior highlights:
  • The prompt instructs the LLM to write a brief summary (bullets or short sentences) emphasizing goals, key decisions, outcomes, blockers, and next steps.
  • Immediately after writing the summary, the LLM must call record_conversation_summary with the resolved parameters, using the server-injected date to avoid ambiguity.
  • Returns: list of messages like [{"role": "system"|"user", "content": "..."}] suitable for your LLM client.
  • See code: seev.prompts.conversation_summary_prompt.

Slash command usage

  • Syntax: /conversation_summary <date|""> <title|""> <conversation_id|""> <inputs|"">
  • Positional arguments, left-to-right. You may omit trailing optional arguments entirely.
  • Use an empty quoted string "" as a placeholder to skip an earlier argument (e.g., blank date means "use today").
  • Quoting: wrap strings with spaces in quotes. Double or single quotes are both allowed.
  • Spacing: use a single space between tokens.

Examples (today is 2025-10-26): - /conversation_summary "" "Cache rollout" "" "Finished caching layer; decided on Redis; add integration tests next" - /conversation_summary 2025-10-26 "Cache rollout" "" "Finished caching layer; decided on Redis; add integration tests next" - /conversation_summary 2025-10-26 "" 42 "Short update"

Rules your assistant should follow (already encoded in the prompt): - Make exactly one tool call to record_conversation_summary. - If conversation_id is provided, ignore title derivation. - If both conversation_id and title are omitted (i.e., both placeholders are empty), derive a short title from the first sentence (≤ 80 chars).