Two agents, one working directory, and a merge nobody asked for
Parallel agents editing one checkout interleave their changes. Isolation costs disk, not correctness.
agents
46 posts
Your prompt is a config file, and it needs the same discipline
Untracked prompt edits are untracked behaviour changes.
Context is a budget, and what you load first wins
Loading everything relevant is not a strategy. Measure tokens per decision.
The retrieval worked and returned the wrong file
Semantic similarity is not relevance. A test file and its subject are near-identical in embedding space.
The agent fixed the test, not the bug, and the test was right
Given a failing assertion, weakening it is the shortest path to green.
The eval that passed because the function returned a plausible stub
An LLM judge scoring output shape rather than behaviour will approve a stub.
An agent loop without a terminal condition is a budget with no bottom
Declare the halt classes before the loop runs. 'Until it looks done' is not a condition.
Your agent's tool output is untrusted input
A file the agent reads can contain instructions. The fix has the same shape as SQL injection's.
A ceiling in a conditional is one someone deletes
A budget expressed as a value can be read, derived from and audited. Buried in a conditional it can only be obeyed or removed, and it gets removed.
A command that checks its own contract
A command's promises decay silently through ordinary edits. Per-command assertions plus a fixture that removes one guarantee is what makes the decay loud.
Not every step needs the expensive model
Tier selection is not a preference, it is a function of three facts about the task. Two platform constraints rule out most of the combinations you would guess.
Loading every rule up front is how none of them apply
An always-loaded rulebook competes with itself for attention. Matching the tool target and injecting only what fires turns 588 rules into the twelve that apply.