2026-01-18 - Create autonomous end-of-session dev diary prompt template

Timeline

• (Context) Needed a portable prompt any AI can run from any working directory, while matching ADDEG’s compiled Markdown output exactly.
• (Action) Surveyed ADDEG’s template, compiler, and config rules to extract the exact required headings, frontmatter format, and bullet rendering behavior.
• (Action) Generated real compiled outputs locally to validate spacing, Boundary Reminder formatting, and section order against actual ADDEG output.
• (Action) Wrote a strict prompt template that enforces output to CWD, exact formatting, topic allowlist, and excerpt clamping rules.
• (Action) Refined the prompt to stop agents from asking questionnaires: require autonomous inference of date/title/topics/events from observed work, with optional overrides only if the user provides them.
• (Action) Added a required completion line (Wrote: <absolute-path-to-file>) so the agent always reports where the new dev-diary entry was saved.
• (Action) Renamed the prompt template file to a descriptive, end-of-work-session name.
• (Observation) Time inference is often unreliable; the prompt now omits times unless they can be confidently derived from visible artifacts.
• (Observation) Small implementation details (PowerShell quoting, tool rename semantics) can derail automation unless the prompt is explicit.
• (Open Thread) Pressure-test the prompt with multiple external AIs to confirm they follow “no questionnaire” and still match ADDEG formatting.

Context

• Needed a reusable end-of-session instruction set that produces a dev diary entry indistinguishable from ADDEG compiled output.
• Target constraint: strict topic allowlist + exact frontmatter/headings/bullets/Boundary Reminder formatting.
• Environment reality: different agents run in different CWDs and often cannot reliably infer timestamps.

Actions

• Located the authoritative formatting sources: core scripts and docs, and the BFD preset rules.
• Generated sample compiled entries to confirm real output formatting and edge cases (spacing, headings, trailing newline).
• Authored a standalone prompt file that enforces: CWD output, filename slug rules, frontmatter formatting, excerpt algorithm, and strict topic allowlist.
• Updated the prompt to require autonomous synthesis from session artifacts (diffs/files/tests/decisions) instead of asking the user for structured inputs.
• Added an explicit “Wrote: ” requirement and renamed the prompt file for end-of-session use.

Observations

• ADDEG formatting is simple but brittle: exact spacing (including the two spaces in the Boundary Reminder line) and consistent headings matter.
• Excerpt quality depends heavily on the first Action item; keeping that first Action sentence concise avoids over-truncation.
• Timestamp precision isn’t consistently available across environments, so “omit times unless confident” is safer than forcing HH:MM.

Open Threads

• Run a quick cross-agent validation: confirm the prompt yields identical structure across at least 2–3 different AI providers.
• Decide whether end-of-session entries should default to draft: false universally or stay true until human review.

Boundary Reminder: Seeds. No maintenance. No roadmap.