Skip to content

Repository files navigation

Engram

Engram turns a person (or a vague need) into a runnable thinking-advisor skill — mental models, heuristics, expression DNA, honest boundaries — plus memory traces and Stage-8 stakes. Soft immersion is opt-in only.

Version 3.0.0
License MIT (LICENSE)

v3 purity. Distill-only northstar: canonical pack path is engram/ (matching GitHub). Scripts own deterministic structure gates (quality_check, test_smoke, doctor/phase1/ax); agents own judgment (fidelity-scorecard, Stage 8 stakes, synthesis). No audit stubs; no vendor branding in load-bearing docs.

Happy path

User says distill <name>; agent owns scripts. User consents only for host installs and checkpoints 1.5/2.5 (approve | revise | stop). See references/happy-path.md.

Install

Place this tree where your agent loads skills:

<skills-root>/engram/

<skills-root> is your agent's skills directory — whatever path that is on the host. Engram never assumes a vendor home folder. Install name is engram only (not a versioned folder name).

  1. Read CONSTITUTION.md, then SKILL.md.
  2. Open references/ files as the phase table in SKILL.md directs.
  3. Write distilled outputs to engrams/<slug>/.
python3 scripts/engram_doctor.py --engram path/to/engrams/<slug>
python3 scripts/phase1_gate.py --engram path/to/engrams/<slug>
python3 scripts/ax_gate.py --engram path/to/engrams/<slug>
python3 scripts/merge_research.py --engram path/to/engrams/<slug>
python3 scripts/quality_check.py --engram path/to/engrams/<slug>
python3 scripts/test_smoke.py   # pack regression vs examples/minimal-fixture

Prefer --engram <dir> (canonical). Positional <dir> / <SKILL.md> still accepted for backward compatibility.

Nested helpers (Phase 1)

Shipped under helpers/ — skill-within-skill:

Folder Job
gemini-video local video → transcript (GEMINI_API_KEY)
web-article-reader full article → markdown
agent-reach multi-platform gather
deep-research structured deep research
pdf PDF/DOCX/EPUB → markdown (MIT huashu-md-html)
doctor preflight + troubleshooting wizards (scripts/engram_doctor.py)

See helpers/README.md and helpers/THIRD_PARTY.md.

Phase 1 hard-binds these into every agent spawn (Helper block + per-agent routing in SKILL.md / references/info-gathering-skills.md).

What ships in this pack

SKILL.md            # Distiller
CONSTITUTION.md     # Non-negotiables
helpers/            # Nested Phase-1 skills (video, article, reach, research, pdf, doctor)
references/         # Phase refs (template, scorecard, scenarios, …)
scripts/            # doctor, phase1_gate, ax_gate, merge_research, quality_check, test_smoke, captions
examples/           # minimal-fixture (smoke / onboarding — not a real subject)
LICENSE VERSION CHANGELOG.md README.md

Intentional cuts

ID Cut
I1 Books only from user-supplied or otherwise legal access
I2 Write path engrams/<slug>/
I3 Engram attribution on shipped person-skills
I4 Non-git version-check skip when the install is not a clone

Not in this repo

  • Distilled person skills (publish separately)
  • Promo / marketing assets

Contributing

Prefer process fixes with receipts. Keep intentional cuts documented.

About

Engram — distill a person into a runnable thinking-advisor skill.

Resources

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages