prog-log

Prog Log Skill

Synchronize capability memory from incremental Git history and persist accepted items through project_memory.py.

Core Responsibilities

1. Read current project memory and last_synced_commit.
2. Collect incremental commits and summarize them for capability extraction.
3. Use Claude reasoning to generate candidate capability JSON.
4. Ask for batch confirmation with index syntax (1,3,5-7).
5. Persist accepted and rejected candidates using project_memory.py.
6. Report inserted/deduped/rejected counts.

Data Layer Command

Use this script for all persistence:

plugins/progress-tracker/prog memory

Main Flow

Step 1: Load Current Memory

Read current memory:

plugins/progress-tracker/prog memory read

Extract:

• last_synced_commit
• existing capabilities for context

Step 2: Collect Incremental Commits

Derive commit range:

• If last_synced_commit exists: commits in last_synced_commit..HEAD
• Else: use recent project history (oldest available to HEAD)

Collect concise commit summaries (hash + subject + touched files) for Claude reasoning.

Step 3: Generate Candidates (Claude Reasoning)

Produce JSON array only. Each candidate must include:

• title
• summary
• tags
• source_commit
• confidence

If parsing fails or candidates are empty:

• report no candidates
• stop without writing

Step 4: Show Numbered Candidate List

Display all candidates with:

• index
• title
• confidence
• source commit

Ask user for accepted indexes:

• supports 1,3,5-7
• empty input means reject all

Step 5: Parse Selection

Parse selection with:

plugins/progress-tracker/prog memory parse-selection --selection "<selection>" --total <N>

On invalid selection:

• show parser error
• ask once again for corrected input

Step 6: Persist Accepted / Rejected

Build accepted_candidates and rejected_candidates by parsed indexes.

Batch write accepted:

plugins/progress-tracker/prog memory batch-upsert \
  --payload-json '<accepted_json_array>' \
  --sync-meta-json '<sync_meta_json>'

sync_meta_json should include:

• sync_id
• commit_range
• last_synced_commit (new HEAD hash)
• rejected_count

Persist rejected fingerprints:

plugins/progress-tracker/prog memory register-rejections \
  --payload-json '<rejected_json_array>' \
  --sync-id '<sync_id>'

Step 7: Final Report

Always report:

1. commit range used
2. total candidates
3. accepted count
4. inserted vs deduped count
5. rejected count
6. new last_synced_commit

Behavior Guarantees

• No semantic algorithm code in data layer.
• Writes are idempotent by fingerprint.
• Sync history and rejected list are retention-limited by script.
• Failure in writing should be reported with actionable error details.