Persistent research tracking for people, brands, and topics. This skill wraps a bundled research engine derived from last30days and saves each run into a dated Markdown timeline.

The bundled research engine can research a topic deeply, but the output is ephemeral. Once the session ends, you lose the accumulated context. last365days adds persistence on top of that workflow.

Each run appends a dated section to a single Markdown file for that person or topic. Over time, that file becomes a timeline you can revisit, diff, export, and extend.

This repo is runtime-agnostic at the file level and ships its own research engine.

last365days is standalone. It vendors scripts/last30days.py plus its supporting library, so users do not need a separate last30days install.

The bundled engine keeps Bird/X behavior inside this repo. Output, cache, and config paths default to last365days namespaces:

• ~/.local/share/last365days/out
• ~/.cache/last365days
• ~/.config/last365days

To install globally:

npx skills add zaydiscold/last365days-skill@last365days -g -y

To install for one agent only:

npx skills add zaydiscold/last365days-skill@last365days -y

By default, saved research goes to ~/Desktop/last365days/. Override that directory with LAST365DAYS_DIR.

persist.py reads stats from ~/.local/share/last365days/out/report.json by default. Override that location with LAST365DAYS_OUTPUT_DIR or the --report-path flag.

The main skill entry point is /last365days.

/last365days kanye west
/last365days AI video tools

Run it again later on the same topic and the skill appends a new dated entry:

/last365days kanye west

If the match is exact or high confidence, the skill appends automatically. If the match is only medium confidence, the skill now asks the user to confirm whether it should reuse the existing file or create a new one.

To browse existing profiles:

/last365days

last365days/scripts/persist.py is the persistence layer behind the skill. You can run it directly when you want diagnostics or structured exports.

These commands are the ones the skill depends on during normal use.

python3 last365days/scripts/persist.py list
python3 last365days/scripts/persist.py match "Kanye West"
python3 last365days/scripts/persist.py history "kanye-west"
python3 last365days/scripts/persist.py read "kanye-west"
python3 last365days/scripts/persist.py search "founder mode"
python3 last365days/scripts/persist.py slugify "Kanye West"

These commands improve trust, debugging, and operability.

python3 last365days/scripts/persist.py doctor
python3 last365days/scripts/persist.py diff "kanye-west" 2026-03-05 2026-04-05
python3 last365days/scripts/persist.py export "kanye-west" --format json
python3 last365days/scripts/persist.py export --all --format csv

doctor checks:

• whether the research directory exists or can be created
• whether the last365days output directory is available
• whether report.json is readable and shaped like a valid report
• whether the bundled research engine is present
• whether optional qmd indexing is available

diff compares two saved date blocks from the same profile using a deterministic unified diff.

export lets you move data out of the Markdown files without hand parsing:

• export <slug> --format md|json|csv
• export --all --format json|csv

The runtime behavior is split cleanly between the skill instructions and the Python helper.

/last365days <topic>
        │
        ▼
SKILL.md
  1. List existing profiles
  2. Match against existing files
  3. Ask for confirmation on medium-confidence matches
  4. Run bundled research
  5. Synthesize and compare against history
  6. Persist the new entry
        │
        ▼
persist.py
  - appends Markdown entries
  - reads report.json for stats
  - exposes history/search/diff/export/doctor commands

Saved files live in ~/Desktop/last365days/ by default:

~/Desktop/last365days/
├── kanye-west.md
├── taylor-swift.md
├── ai-video-tools.md
└── ...

The Markdown format is documented in last365days/references/file-format.md.

If you need to understand where to look before changing something, start here.

• last365days/SKILL.md: runtime workflow, matching rules, and user-facing research behavior
• last365days/scripts/last30days.py: bundled research engine entry point
• last365days/scripts/lib/: bundled research engine support modules
• last365days/scripts/persist.py: persistence CLI, storage parsing, exports, diffs, and diagnostics
• last365days/references/file-format.md: profile file structure and same-day update format
• last365days/references/operations.md: browse, export, diff, and troubleshooting flows that are intentionally kept out of the main skill path
• tests/test_persist.py: CLI regression coverage for slugify, matching, history, search, doctor, diff, and export
• .github/workflows/ci.yml: syntax check plus test execution on push and pull request

This repo now includes a minimal built-in Python test suite and a GitHub Actions workflow.

Run everything locally with:

python3 -m compileall last365days tests
python3 -m unittest discover -s tests -v

CI runs the same two checks on every push to main and on pull requests.

The current design stays intentionally lean. A few limits are deliberate.

• Medium-confidence matches still rely on user confirmation rather than richer identity metadata.
• diff compares whole dated Markdown blocks. It does not produce semantic summaries.
• export --all --format md is not supported.
• Source stats depend on a compatible report.json produced by the bundled engine.
• There are no destructive workflows yet for merge, delete-entry, or delete-profile.
• File writes are simple append operations. There is no broad locking or atomic rewrite strategy yet.

These items are intentionally deferred because they add more risk than the v1.1 changes.

• merge, delete-entry, and delete-profile with backup and confirmation
• profile metadata blocks such as aliases, canonical handles, and timestamps
• broader locking and atomic rewrite strategy for concurrent edits

• trimmed vendored Bird dependency files that are not needed at runtime (.d.ts and source maps)
• re-verified the bundled engine, diagnostics, and test suite after the trim

• vendored the last30days research engine into this repo so last365days works standalone
• moved default output, cache, and config paths to last365days namespaces
• updated docs and diagnostics to validate the bundled engine instead of an external install

• clarified X/Twitter behavior and Bird troubleshooting

• added doctor diagnostics for path, dependency, and report.json validation
• added diff <slug> <date1> <date2> for deterministic profile comparisons
• added export support for Markdown, JSON, and CSV outputs
• added a test suite for core CLI behavior
• added GitHub Actions CI for syntax and tests
• changed medium-confidence matches to require explicit user confirmation
• expanded documentation with a repo map, command reference, and known limits

• added Anthropic skill guide metadata and compatibility frontmatter
• moved file format details into references/file-format.md
• added troubleshooting and better trigger phrases
• added search plus CLI path overrides for persist.py
• improved Unicode slug handling and deterministic fallback slugs
• guarded the qmd after-hook so it only runs when available

• shipped persistent per-person and per-topic Markdown timelines
• wrapped last30days research output in a reusable profile model
• added same-day duplicate handling and automatic source stats

MIT