Epistemic Deconstructor
Six-phase Bayesian reverse-engineering protocol with disk-persisted state and gated phase transitions.
Test report
- Verdict
- Works with setup
- Score
- Tested
- Jul 15, 2026
- Environment
- Claude Code 2.x (agent harness)
- Upstream re-checked
- Aug 10, 2026 · 127dced
Upstream indeed moved from v7.16.4 to v7.16.7 (four releases on 2026-07-16), but the diff shows that exactly two lines changed in SKILL.md itself: the version number in the header and an added link to warning-reception.md in phase 0.7; changes went into agent files, not the skill body. The exact problem we documented was not fixed by the author: the section on `make sync-skill` is verbatim the same as in v7.16.4, and a live run confirmed this — after a clean `git clone` + `cp -r src`, the ~/.claude/agents directory does not exist at all, meaning all 14 specialists are undispatchable, yet `make sync-skill` places 15 agent files there. Checked the gates properly in real sessions: `advance` without a declared tier is rejected with a clear error, `skip 0.3` without `domain_familiarity: high` is denied, `set-phase --force-state` actually writes ADMIN-OVERRIDE to decisions.md. Found a new discrepancy between docs and code: SKILL.md states 'Every other skip is refused', but `skip 2` is accepted and logged by the script (the phase cursor does not move) — the protocol is not bypassed, but the promise is stronger than reality. Stage 0 clean: read the full body + went through all 15 scripts, no network or exfiltration, the only `exec()` in simulator.py runs ODE-code through an AST-validator and restricted builtins. The setup verdict remains: the manual `make sync-skill` step is still there.
Scored on four weighted criteria — install, triggering, output vs. baseline, docs. How scoring works
- Installs cleanly 4/5
- Triggers reliably 5/5
- Output vs. baseline 6/10
- Docs & honesty 3/5
What Epistemic Deconstructor does
Runs a structured, gate-enforced scientific protocol for reverse-engineering unknown systems (black-box APIs, security side-channels, competitive intelligence, forecasting) with Bayesian hypothesis tracking persisted to disk. Triggers on black-box analysis, security/side-channel investigation, competitive intelligence, forensics, or predictive-model requests; the full multi-agent experience needs an extra install step beyond a plain skill copy.
How to install Epistemic Deconstructor
git clone https://github.com/NikolasMarkou/epistemic-deconstructor
cd epistemic-deconstructor && make sync-skill # ставит скил в ~/.claude/skills/ И 15 агентов в ~/.claude/agents/; затем перезапустить сессию
Skills live in ~/.claude/skills/ (global) or .claude/skills/
(per-project). Restart Claude Code after installing.
Commands — how to trigger Epistemic Deconstructor
-
/epistemic-deconstructorSix-phase Bayesian reverse-engineering protocol with disk-persisted state and gated phase transitions.
It also activates on plain-language prompts like these:
-
Figure out if this rate limiter is token-bucket or fixed-window -
Reverse-engineer this API's undocumented behavior from observations -
Investigate this side-channel timing pattern and rank hypotheses
Frequently asked questions
- Is the Epistemic Deconstructor skill free?
- Yes. The skill itself is free from NikolasMarkou/epistemic-deconstructor. SkillProof publishes the install command and an independent test verdict at no cost.
- Does Epistemic Deconstructor work with Claude Code?
- We tested it with Claude Code 2.x (agent harness) on Jul 15, 2026. Verdict: Works with setup. Upstream indeed moved from v7.16.4 to v7.16.7 (four releases on 2026-07-16), but the diff shows that exactly two lines changed in SKILL.md itself: the version number in the header and an added link to warning-reception.md in phase 0.7; changes went into agent files, not the skill body. The exact problem we documented was not fixed by the author: the section on `make sync-skill` is verbatim the same as in v7.16.4, and a live run confirmed this — after a clean `git clone` + `cp -r src`, the ~/.claude/agents directory does not exist at all, meaning all 14 specialists are undispatchable, yet `make sync-skill` places 15 agent files there. Checked the gates properly in real sessions: `advance` without a declared tier is rejected with a clear error, `skip 0.3` without `domain_familiarity: high` is denied, `set-phase --force-state` actually writes ADMIN-OVERRIDE to decisions.md. Found a new discrepancy between docs and code: SKILL.md states 'Every other skip is refused', but `skip 2` is accepted and logged by the script (the phase cursor does not move) — the protocol is not bypassed, but the promise is stronger than reality. Stage 0 clean: read the full body + went through all 15 scripts, no network or exfiltration, the only `exec()` in simulator.py runs ODE-code through an AST-validator and restricted builtins. The setup verdict remains: the manual `make sync-skill` step is still there.
- What is the Epistemic Deconstructor SkillProof Score?
- 7.2/10 — installs cleanly 4/5, triggers reliably 5/5, output vs. baseline 6/10, docs & honesty 3/5.
- How do I install Epistemic Deconstructor?
- Copy the install command from this page, run it in your terminal, and restart Claude Code. Skills live in ~/.claude/skills/ (global) or .claude/skills/ inside a project.
- Can I use Epistemic Deconstructor with Cursor, Copilot, Gemini CLI, Codex or other AI tools?
- The SKILL.md format is native to Claude (Claude Code, Desktop, claude.ai). The instructions inside adapt to other assistants: Cursor rules, GitHub Copilot instructions, Windsurf rules, Custom GPTs, AGENTS.md for OpenAI Codex, and GEMINI.md for Google Gemini CLI — our conversion guides cover each, and the free converter on the tools page does the wrapping for you.