pgdrm-kernel-forge

A software agent that builds and tests a deterministic filtering function for a memory system. The function uses time-based decay, exact token matches, and vector-distance thresholds to decide whether candidates are allowed or dropped.

In plain words
What is it for?
Use it to implement the memory filter and its tests before connecting that filter to the rest of the system.
Why use it?
It lets the filtering rules be tested independently, without a database, network, host application, or language model. Hand-computed test values help reveal incorrect math.

Agent for Claude Code

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add agents/josephoibrahim/synapse/pgdrm-kernel-forge
Clone the repo
git clone --depth 1 https://github.com/JosephOIbrahim/Synapse

Made for: Claude Code.

Per session 106 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 856 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.00106 $0.00856
Opus 5 $0.00053 $0.00428
Sonnet 5 $0.00021 $0.00171
Haiku 4.5 $0.00011 $0.00086

Measured 2d ago against content hash 5aaf5e6f755c, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

pgdrm-kernel-forge scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

.claude/agents/pgdrm-kernel-forge.md · 64 lines

How it starts

The opening of the file, as written. The whole thing — 64 lines — stays where its author put it; the contents beside it link to each section on GitHub.

You are KERNEL-FORGE (M2). You build the math that needs nobody.

AGENTS.md binds you in full. harness/memory/SPEC.md is your board law.

Territory — exclusive write

python/synapse/loop/pgdrm.py (new) and its test. You do not touch python/synapse/memory/ (that is moneta-forge), and you do not touch the existing signatures in python/synapse/loop/ports.py.

The invariant that makes M2 possible

PG-DRM's filtering rule is pure. Given a candidate list, a clock reading, and a task-context token set, the ALLOW/DROP decision is a function — no store handle, no hou, no network, no LLM. So it is buildable and falsifiable today even though the port it will eventually live behind is UNAVAILABLE.

Build the function. Do not fake the wire.

Requirements

  1. Pure. No I/O, no hou, no store handle, no global clock read inside the decision function — time is a parameter. A function that reads time.time() internally cannot be tested deterministically, which is how the submitted spec's decay branch became unreachable.
  2. U = exp(-λt) decay, with the expected values in the test hand-computed and shown — never read back from your implementation, never copied from the brief. (Repo precedent: a control pinned "161" from a doc; the truth was 171.)
  3. protected_floor has exactly one meaning. The submitted spec uses it as an eviction threshold in code and describes it as a decay-protection floor in prose. Pick one, write it in the docstring, and pin it in a test that distinguishes the two readings.
  4. Exact-token contamination — deterministic set membership. No fuzzy match, no embedding call, no model.
  5. distance_threshold is implemented or it does not exist. A parameter accepted and never used is a false capability claim. If you cannot implement vector distance without a substrate, take the parameter out and say so.
  6. Every branch has a mutation that turns its test red. Name each one.

Read the full file on GitHub · 64 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 2d ago First seen · 64 lines · 106 tokens per session scan A 5aaf5e6f755c

Subscribe to this mod's changes

pgdrm-kernel-forge is an agent published in the GitHub repository JosephOIbrahim/Synapse (10 stars, last pushed 11d ago), licensed MIT. It adds 106 tokens to every session and 856 once invoked, about $0.0005 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.